Essays on Top Scholars: a Scientometrics Approach

ESSAYS ON TOP SCHOLARS: A SCIENTOMETRICS APPROACH

Ho Fai Chan

BBus (Econ), MRes (Econ)

Principal Supervisor: Prof. Benno Torgler

Associate Supervisor: Dr. Markus Schaffner

Submitted in fulfilment of the requirements for

Doctor of Philosophy (Economics)

School of Economics and Finance

Queensland University of Technology

2017

Abstract

This thesis contributes to the scientometrics literature by providing empirical evidence on a number of aspects in academia focusing on the most prominent scholars. This thesis consists of eight studies. The first two studies explore the recognition pattern through major institutionalised awards before and after the Nobel Prize and the implications of educational and methodological background on future recognitions. The third and fourth studies concerns the collaboration pattern and performance of Nobel laureates, where I examine the existence of structural break on collaboration behaviour at the reception year of the Prize, I also explore the efficiency of repeated collaborations. The fifth and sixth studies explore the relationship between academic success and societal impact; and investigate scholars can capitalise them in the public speaking market. The seventh study use up to 100 years of economic research to analyse the time effect on citation bias due to status prestige. The last study draws on data derived from millions of English books published since the 1800s to visualise the emergence of great scientific minds.

Keywords

Nobel Prize; Scientific collaboration; Matthew effect; External influence;

Symbolic capital; Societal impact; Awards; Great minds; Recognition;

Speaking fee; Citation bias

Acknowledgments

First of all, I would like to express my sincere thanks to my supervisor, Prof. Benno Torgler, for the immeasurable amount of patient guidance and encouragement he has provided me throughout the time as his student. Without his support, both intellectually and emotionally, this PhD would not have been achievable. I have been extremely lucky to have you as my mentor and I am truly grateful. I must also take this chance to say thanks and apologise to the Torgler family, for the endless hours I have taken Benno away from you.

I am also profoundly indebted to my secondary supervisor Dr. Markus Schaffner, who have always assisted me with patience whenever I sought your help, even with trivial problems. Thank you for devoting the time in developing and enriching my technical knowledge and skills over the years. Not only do they have proven to be very useful but I also very much enjoyed the time learning and practicing them.

Special thanks must be made to my friends and colleagues at QUT, particularly Marco, Yang, Steve, Naomi, Juliana, Tony, Zili, Ann-Kathrin, Dave, Uwe, Azhar, Jonas, Susan, Sam, Osei, Poli, Justin, and Nate for making my time as a graduate student most enjoyable.

I would like to say a heartfelt thank you to my parents and brothers for supporting me during this whole period of studying abroad and being away. And to Kellie, thank you for being there for me.

I am also thankful to my co-authors for their valuable input and hard work into each of the studies of this thesis, in particular to Ali Önder, Bruno Frey and Jana Gallus, for such efficient and fruitful collaborations. I also gratefully acknowledge Alison Macintyre for her editing and proofreading advice.

Thanks are due to the anonymous referees for advice and suggestions to various studies included in this thesis as well as the outstanding help from Marco for the study in Chapter two. Financial support from the Australian Research Council (FT110100463) is also acknowledged. Lastly, I also gratefully acknowledge the staff of the administration office in the School of Economics and Finance and the Research Support office for providing me all necessary assistance to this graduate study.

iii

Table of Contents

List of Figures ix

List of Tables xi

List of Publications xiii

Chapter 1 General Introduction 1

Outline of the Thesis 6

Chapter 2 Awards Before and After the Nobel Prize: A Matthew Effect and/or a Ticket to One's Own Funeral? 9

Abstract 9

Introduction 9

Theoretical considerations 11 Matthew effect in science 11 Possible adverse effects of awards 14 What about age? 16 Methods 17 Data collection 17 Statistical analysis 18 Results 19

Discussion 27

Appendix 29

Chapter 3 The Implications of Educational and Methodological Background for the Career Success of Nobel Laureates: An Investigation of Major Awards 31

Abstract 31

Introduction 31

Method and data 35

Results 38 Descriptive analysis 38 Multivariate analysis 47 Conclusions 52

Chapter 4 Do Nobel Laureates Change Their Patterns Of Collaboration Following Prize Reception? 55

Abstract 55

Introduction 55

A descriptive analysis of collaboration trends 58

Multivariate analysis 71

Conclusions 79

Chapter 5 The First Cut is the Deepest: Repeated Interactions of Co-Authorship and Academic Productivity in Nobel Laureate Teams 81

Abstract 81

Introduction 81

Data and descriptive analysis 82

Two-stage estimation and discussion of results 89

Conclusion 93

Appendix 95

Chapter 6 External Influence as an Indicator of Scholarly Importance 103

Abstract 103

Introduction 103

Capturing the external influence of economics and economists 106 Reflections of the influence via markets 107 Reflections of the influence via persons 108 Reflections of the influence via outside markets 109 Methodological approach 110

Results and discussion 114

Conclusions 137

Appendix 139

Chapter 7 Do The Best Scholars Attract the Highest Speaking Fees? An Exploration of Internal and External Influence 145

Abstract 145

Introduction 146

Scholarly impact 148 Reward mechanism 149 Social effects 150 Data 152

Speaking fees 152 External influence 156 Descriptive analysis 159 Internal influence 161 Estimation results 164

Discussion and conclusions 173

Appendix 175

Chapter 8 The Inner Quality of an Article: Will Time Tell? 179

Abstract 179

Introduction 180

Data 183

Descriptive analysis 185

Multivariate analysis 193

Conclusions 197

Appendix 198

Chapter 9 Do Great Minds Appear in Batches? 205

Abstract 205

Introduction 205

Measuring trends in great minds 208

Materials and methods 210

Results 214

Conclusions 223

Appendix 224

Chapter 10 Summary and Conclusions 227

Summary of findings 227

Shortcomings 231 Chapter 2 and 3: Recognition cycle with academic awards 232 Chapter 4 and 5: Collaboration pattern and productivity 232 Chapter 6 and 7: External and internal influence 233 Chapter 8: Quality of economic research in the long run 234 Chapter 9: Great minds 234 Direction for future research 235

Bibliography 237

vii

Appendices: Statements of Authors Contributions 259

viii

List of Figures

Figure 2.1 Number of major awards before and after the Nobel Prize 20 Figure 2.2 Share of major awards within different fields (5-year moving average) 20 Figure 2.3 Publication performance pre and post Nobel Prize 24 Figure 3.1 Awards and countries’ educational background 40 Figure 3.2 Nobel laureates who obtained the Nobel Prize before and after WWII 40 Figure 3.3 Institutions and award success 43 Figure 3.4 Institutions and award success before and after WWII 43 Figure 3.5 Empirical versus theoretical orientation 45 Figure 3.6 Pre- and post-award difference between sub-fields 46 Figure 4.1 Average of Nobelists’ new coauthors before and after the Nobel Prize 59 Figure 4.2 Different age cohorts based on the age at prize reception 61 Figure 4.3 Number of new coauthors by field 61 Figure 4.4 New coauthors divided by the number of existing coauthors 63 Figure 4.5 Number of pre-award coauthor dropouts 66 Figure 4.6 Dropout rates for pre-award coauthors 67 Figure 4.7 Dropout rate by age cohort 67 Figure 4.8 Dropout rate by field 68 Figure 4.9 Dropout rates including post-award coauthors 70 Figure 5.1 Arrival of new coauthors by field 83 Figure 5.2 Intensity of cooperation by field 84 Figure 5.3 Citations received by early and late collaborations of laureate-coauthor pairs 87 Figure 5.4 Distribution of the total number of Nobel laureate coauthors 95 Figure 5.5 Intensity of cooperation by field weighted by unequal co-author contribution 96 Figure 5.6 Intensity of cooperation by field weighted by equal co-author contribution 96 Figure 5.7 Citations received by early and late collaborations of laureate-coauthor pairs weighted by unequal co-author contribution 97 Figure 5.8 Citations received by early and late collaborations of laureate-coauthor pairs weighted by equal co-author contribution 98 Figure 5.9 Citations received by first and second half of all collaborations of laureate-coauthor pairs 99 Figure 6.1 Lorenz curves for Google and Bing page counts 115 Figure 6.2 Rank order differences across lists, by either internal or external influence measure 117 Figure 6.3 Google trends for Nobel laureates before and after the Nobel Prize 127

Figure 6.4 Google trends for John Bates Clark Medallists before and after the award 127 Figure 6.5 Rank order differences compared to Aguinis et al. (2012) 139 Figure 7.1 Distribution of minimum speaking fees 155 Figure 7.2 Impact inequality 157 Figure 7.3 External prominence and minimum speaking fee 164 Figure 7.4 Speaking fees and external impact in sub-fields 175 Figure 8.1 Citation rank difference over time for authors belonging or not to a top ten university 187 Figure 8.2 Citation rank difference over time in Econometrica, by decade of publication 192 Figure 8.3 Contrasts of predictive margins (by top ten university) 195 Figure 8.4 Citation rank difference over time for authors belonging or not to a top 20 university 198 Figure 8.5 Citation rank difference over time, by decade of publication (top 20 university) 199 Figure 8.6 Citation rank difference over time for authors obtaining a Ph.D. in a top ten university 200 Figure 8.7 Citation rank difference over time, by decade of publication (top ten Ph.D.) 201 Figure 8.8 Contrasts of predictive margins (by top 20 university and Ph.D.) 202 Figure 9.1 Distribution of eminence across fields 212 Figure 9.2 Number of scientists born each year 213 Figure 9.3 Timeline of great minds born 1800–1969 217 Figure 9.4 Eminence development by field based on mean mD 224 Figure 9.5 Eminence development by field based on sum mD 225

List of Tables

Table 2.1 Results of random effects negative binomial regression models 22 Table 2.2 Awards, productivity, and cohorts 26 Table 2.3 Top 20 major awards obtained by Nobel laureates (1901-2000) by field 29 Table 3.1 Educational differences 42 Table 3.2 Educational institutional differences 44 Table 3.3 Theory versus Empirical Orientation 46 Table 3.4 The impact of educational and methodological background on awards using random effects negative binomial model 49 Table 3.5 Institutional Educational Background and Award Success 51 Table 4.1 Chow test for structural breaks and mean comparison t-test: number of new coauthors 60 Table 4.2 Chow test for structural breaks and mean comparison t-test: entry rate 62 Table 4.3 Chow test for structural breaks and mean comparison t-test: number of pre-award coauthor dropouts 65 Table 4.4 Chow test for structural breaks and mean comparison t-test: dropout rates for pre-award coauthors 68 Table 4.5 Chow test for structural breaks and mean comparison t-test: dropout rates for all coauthors 70 Table 4.6 Effects before and after the Nobel Prize 73 Table 4.7 Pre-award collaboration intensity and loyalty 77 Table 5.1 Ratio of early to late citation success 89 Table 5.2 Regression results for the 2SLS 91 Table 5.3 A-index for unequal co-author contributions 100 Table 5.4 First Stage Regression Results for 2SLS 100 Table 5.5 Descriptive statistics of dependent and independent variables employed in 2SLS regression analysis 101 Table 6.1 Correlations between external influence and internal performance (RePEc rankings) 121 Table 6.2 Correlations between external influence and total internal academic impact (Publish or Perish and Web of Knowledge) 124 Table 6.3 Correlation between external influence and prizes and awards received 125 Table 6.4 Determinants of external influence 131 Table 6.5 Determinants of external influence (extended version) 135 Table 6.6 Ranking of economics scholars by average number of web counts 141 Table 6.7 Correlations based on corrected values 143 Table 7.1 Sample size by academic involvement and fields 155 Table 7.2 Academics versus non-academics 160

Table 7.3 Correlation matrix of external influence proxies 161 Table 7.4 Speaking fee and external influence 167 Table 7.5 Academic performance and speaking fees 171 Table 7.6 Correlation Matrix of internal impact proxies (Google Scholar and Scopus) 177 Table 8.1 Mean citation rank difference in Econometrica, by year since publication 189 Table 8.2 Mean citation rank difference in AER, by year since publication 191 Table 8.3 Results of random-effects GLS regression models (top 10 university and Ph.D.) 194 Table 8.4 Institutional Ranking 203 Table 8.5 Results of random-effects GLS regression models (top 20 university and PhD) 204 Table 9.1 Kolmogorov–Smirnov equality-of-distributions test for mD across fields 213 Table 9.2 Non-parametric test for non-randomness 221 Table 9.3 Exploration of different mD percentiles 222

xii

List of Publications

This thesis is presented in the form of published papers. In particular, the thesis is comprised of the following eight manuscripts.

1. Chan, H. F., Gleeson, L., & Torgler, B. (2014). Awards before and after the Nobel Prize: a Matthew effect and/or a ticket to one's own funeral? Research Evaluation, 23(3), pp. 210-220. 2. Chan, H. F. & Torgler, B. (2015). The implications of educational and methodological background for the career success of Nobel Laureates: an investigation of major awards. Scientometrics, 102(1), pp. 847-863. 3. Chan, H. F., Önder, A. S., & Torgler, B. (2015). Do Nobel laureates change their patterns of collaboration following prize reception? Scientometrics, 105(3), pp. 2215-2235. 4. Chan, H. F., Önder, A. S., & Torgler, B. (2016). The first cut is the deepest: Repeated interactions of co-authorship and academic productivity in Nobel laureate teams. Scientometrics, 106(2), pp. 509- 524. 5. Chan, H. F., Frey, B. S., Gallus, J., Schaffner, M., Torgler, B., & Whyte, S. (2016). External influence as an indicator of scholarly importance. CESifo Economic Studies, 62(1), pp. 170-195. 6. Chan, H. F., Frey, B. S., Gallus, J., Schaffner, M., Torgler, B., & Whyte, S. (2014). Do the best scholars attract the highest speaking fees? An exploration of internal and external influence. Scientometrics, 101(1), pp. 793-817. 7. Chan, H. F., Guillot, M., Page, L., & Torgler, B. (2015). The inner quality of an article: Will time tell? Scientometrics, 104(1), pp. 19-41. 8. Chan, H. F. & Torgler, B. (2015). Do great minds appear in batches? Scientometrics, 104(2), pp. 475-488.

Additionally, the following paper was submitted for publication during the course of candidature but does not contribute to the thesis:

9. Ong, D. Chan, H. F., Torgler, B. & Yang, Y. A. (2016). Collaboration incentives: Endogenous selection into single and coauthorships by surname initials in economics and management. Canadian Journal of Economics/Revue canadienne d'économique manuscripts under revision following referees' reports.

xiii

Statement of Original Authorship

The work contained in this thesis has not been previously submitted to meet requirements for an award at this or any other higher education institution. To the best of my knowledge and belief, the thesis contains no material previously published or written by another person except where due reference is made.

QUT Verified Signature

Signature:

Date: 06/03/2017

xiv

Chapter 1 General Introduction

Studying scientists is vital. Scientists and researchers play an important role in society as key driving force behind enriching the collective human knowledge and advancing technological capabilities and innovation. Such intellectual contributions can be viewed as an important form of public good where it links to the eventual progress in social and economic well-being. Developing a better understanding of the mechanisms behind how scientists think and learn, and how they react to and interact with the surrounding environment and people would be beneficial to improve the design of the research systems, and hence, potentially increase the efficiency of scientific knowledge production. In particular, organisations such as government funding bodies and research institutions have great interests to know what sort of scientific regulations and systems work and what doesn’t work and how to improve the institutional structure. Examples of topic of interest are career path of scientists, research performance evaluation, research topics and directions, and resources allocations and reward systems (e.g. hiring, promoting, funding decisions).

Studying the academic labour market also offers a number of advantages in terms of methodology when compared with the traditional labour market setting. First, academia is comparable to a controlled experimental setting in which subjects (academics) have highly homogenous tasks (conduct research), share a similar job profile (academic training and research environment), and also have the same goals (to produce scientific knowledge) as opposed to the highly heterogeneous setting in the usual work force. In particular, the benefit of analysing top scholars is that they comprise a homogeneous sample of highly skilled and successful people. Second, individual performance in the academic labour market setting is highly transparent and easily obtainable compared to the labour market environments. For example, most academics have made their CV publicly available, in which detailed records of personal and academic achievement are recorded (number of publications, grants, and awards, as well as their personal and educational details). In addition, citation databases are also a great source of information where publication and citation data are

documented systematically. This in turn offers a myriad of opportunities to test social and economic theories empirically using the academics’ data. Furthermore, the continual collection of data on academia also allows one to quantify individual scientists’ activities as well as scientific progress over time where theories can be robustly tested not only on a static but also a dynamic level.

As such, the present thesis aims to contribute to the study on prominent scholars by focusing on various social and economic aspects of academia, with the goal of providing additional empirical evidence and filling the void in the current literature. In general, the distribution of scientific productivity among individuals is highly skewed in nature, meaning that a large share of the scientific contributions are generated by a small fraction of scientists, indicating a pattern that follows a power law distribution (Lotka 1926). This phenomena provides the motivation for this thesis to places its focus on groups of scientists that were deemed most prolific and innovative, whose contributions to the scientific knowledge are the greatest. The subjects of interest in this thesis are Nobel laureates; economists who were ranked highest or published in best ranked journals; and the greatest scientific minds of the last 200 years. Specifically, this thesis presents eight individual studies on the topics on recognition patterns, effect of positive status shock, collaboration behaviour, long run citation bias, and the emergence of great scientists. The common theme among these studies is the scientometrics approach of analysing science process using quantitative methods.

One of the key driving forces behind the knowledge production of scientists is the desire for prestige, and for recognition by their peers. Such recognition comes in various forms, one of which is through the use of awards. In fact, the use of awards, prizes and honours in the scientific community has an historic and extensive role compared to other spheres of society1 (Frey and Neckermann 2009; Zuckerman 1992). Yet, it was only recently that research started honing in on the design and the mechanisms of

1 For example, the list of awards in science and technology in Wikipedia is not smaller than other sections, especially when contrasted with the size of the academic labour force relative to others. https://en.wikipedia.org/wiki/List_of_prizes,_medals_and _awards#Science_and_technology.

2 reward systems and incentivising instruments in science (e.g. Frey and Osterloh 2010; Stephan 2012). Many aspects of the award culture within the scientific community still require further examination and exploration, both theoretically and empirically. For example, despite the abundant number and popular use of awards available to almost every field of science, there is not yet a database or systemic study that provides a collection of the overall distribution of academic awards. Such information would allow a complete “mapping” of the recognition path of the members of the scientific society, which then could enable further examination as to the implications of such dynamic of recognition and its relationship with other social aspects of the scientific community. The first two studies of this thesis attempt to provide some empirical evidence upon which future research could build. In Chapter 2, I present the first empirical evidence of the life-cycle allocation of major awards in science, drawing on a large biographical dataset collected on the most esteemed scientists – Nobel laureates of the first century in Physics, Chemistry and Physiology or Medicine. The immediate outcome of this study of award patterns is the common phenomenon of cumulative advantage in science (or the Matthew effect as termed by Merton (1968)), which was evidenced and empirically tested in citation research (e.g. Cole 1970). The theoretical framework of the Matthew effect in science can be described as the continuous feedback loop where the incremental gain in recognition is based on the level of current status as opposed to the performance or inherent quality. Intuitively, many awards share very similar functions as means of recognition, i.e. honour bestowed upon the best performer. The overlapping criteria of awards could lead to such a Matthew effect, as Herbert Simon once said, ‘after a while the criterion for getting an honor is to have been awarded a lot of other honors’. Moreover, research in the last few years has also shown that obtaining prestigious awards in science increases citation and publication performance (e.g. Azoulay et al. 2014; Chan et al. 2013; Borjas and Doran 2015). Yet, to the best of my knowledge, there is no research which provides empirical insights into the phenomenon that awards breed awards. Furthermore, another aspect explored in Chapter 2 is whether the award pattern changes dramatically after the Nobel Prize – the highest accolade in science – is obtained. The implications of this honour are also discussed. Moreover, as one could assume the Nobel laureates are a comparable group, their award pattern should be similar over their scientific career. Thus, in Chapter 3 I explore the whether there is a variation of the award life cycle that

3 could be explained by laureates’ educational background, as this influence on future success is often suggested. Specifically, I explore the differences in the number of major awards obtained by laureates during their entire career by the location and institution of where they received their education and the methodological approach (theoretical or empirical methods) undertaken.

In Chapter 4, I explore the implications of award reception with another important aspect of scientific knowledge production: i.e. research collaboration. Conducting research in solitude is no longer the common practice in modern science; in fact, solo-author research has become a rarity in recent years. On the other hand, the phenomenon of teamwork in science has become the dominant practice due to increasing task difficulty which requires multiple forms of expertise (Katz and Martin 1997; Adams et al. 2005). This has also led to a growth of research into understanding of scientific team and network formation (e.g. Guimera et al. 2005; Milojević 2014), and their productivity and impact compared to single author research (Wuchty et al. 2007). However, I am not aware of any research that has systemically analysed the effect of an exogenous status shock on the coauthorship pattern dynamic. Chapter 4 therefore explores whether there is a positive or negative change in collaboration behaviour after the reception of the Nobel Prize. In particular, using data on the collaborative works of Nobel laureates, I examine and discuss the effect on collaboration intensity and sustainability of which coauthors enter and exit the Nobel laureates’ collaborator pool after the extreme positive status shock.

In addition to the change in collaborative patterns in science, it is also of interest to understand how creativity is linked to the repeated interactions among scientists. A better understanding of the relation between knowledge production efficiency within the longevity of collaborative relationships could aid scientists in decision-making about team collaboration. Hence, Chapter 5 aims to provide some insights on this topic by examining the performance difference of repeated collaborations over time, using a sample of scientific teams comprised of the most prominent researchers (Nobel laureates). The results of the study emphasise the implications of the intensity and long-lasting nature of collaborations.

Measuring and describing research impact has become an important issue around the world. Traditionally, evaluating the level of impact by

4 research study or scientists rely on indicators developed based on internal academic metrics such as impact factors or h-index. However, capturing the external impact of scientists and researchers is increasingly attracting interest (Jensen et al. 2008) and its importance is often emphasised. For example, rating research based on their engagement or influence to the wider public, through the economic, social or environmental sphere. In particular, the LSE Public Policy Group has dedicated a handbook solely to advising how social scientists can maximise their research impact. As such, in Chapter 6, I provide a detailed discussion on how the societal impact of economics and economists could be captured. I employ an innovative approach, enabled by Internet search engines, to proxy the societal influence of top economists and compare them with their internal prominence within academia using correlation analysis.

In addition, while the internal remuneration to academics is often heavily based on their internal academic productivity (e.g. how many publications per year and where are they published), the determinants of the remuneration process to academics engaging in external activities remains largely unknown. This is largely due to the fact that payment records for external work such as consultancy, report writing or organising research workshops to the private sector are rarely documented. Such information could also be very hard to standardise for quantitative analysis. Nonetheless, in Chapter 7 I show first evidence on the determinants of remuneration to scientists in the marketplace. Specifically, drawing on a novel dataset of public speaking fees required by scientists, it provides the first empirical analysis on how academics could capitalise their internal and external prominence in the speaking market.

The use of citation count to measure recognition is not free of problems. Despite its frequent use as an evaluative tool for individual scientists’ scholarly contribution and performance, numerous studies in various disciplines have demonstrated that the decision to cite an article could be based on subjective factors, and prejudices are often involved in the process. Such citation bias could be defined as the decision to cite (or not cite) an article based on specific factors or attributes other than the quality or the work itself, compared to another article of the same quality but without those attributes. Examples of these attributes are authors’ gender, race, country of residence, work affiliation, significance of results, journal prestige etc., where

the citation bias led by these factors are quickly observable soon after the article being published. Yet, one might ask the following question: given enough time, would science be able to self-correct, thereby eliminating such biases, or would this phenomenon become more prevalent over time? Chapter 8 thus aims to address this question by comparing the yearly citation performance of up to 80 years of articles in published in Econometrica and American Economic Review. In particular, I examine the case of citation bias arising from the prestige of the authors’ work affiliation and the institution where the doctoral education was received. I assume articles published in the same journal at the same time (same issue) should have similar quality as they went through the same editorial process (same managing editor, editorial board members etc.) and this should be reflected in its citation performance. This study thus empirically test the question of whether the quality of a scientific contribution will survive the test of time in a setting analogous to a controlled experiment.

In the final study of the thesis, I go further back in time to the beginning of the 19th century, and address a question that has garnered much fascination: Do great scientific minds appear in cycles, or do they grow in the same fashion as technological advancement? Many have theorised that great minds emerge because of their surrounding environment. For instance, Simonton (1975) attributes sociocultural and setting such as zeitgeist, political fragmentation, war, civil disturbances, and political instability to important environmental factors in shaping future genius. Drawing on inspiration from good mentors and great peers through intensive interactions is also deemed extremely influential. This in turn led to the hypothesis that the number of geniuses throughout history should not be random and a clear pattern could be observed. This, in the last study of this thesis, I will provide some insights to this fascinating question using a novel dataset derived from 4% of all books in print over the a period of 200 years. Through the use of the ‘literal frame’ of 5630 scientists as a proxy of creativity, I will visualise the emergence of scientific great minds since the beginning of the 19th century and test whether there exists a cyclical pattern of emergence or whether it is pure coincidence.

Outline of the Thesis

In the following I present the eight essays comprising this PhD study in individual chapters. Each chapter is self-contained and thus the relevant

6 literature is independently presented within each chapter. Due to the requirement for thesis by publications, a few adjustments from the published versions have been made for ease of the reader. These changes include altering the spelling format from American to British, renumbering and reformatting of tables and figures and combining all references into a single list for the entirety of the thesis.

First, I present the two explorative studies on honorific award lifecycle of individual eminent scientists. Chapter 2, titled Awards Before and after the Nobel Prize: A Matthew effect and/or a ticket to one's own funeral?, is a joint work with Laura Gleeson and Benno Torgler and published in Research Evaluation. Chapter 3 presents the study in collaboration with Benno Torgler (published in Scientometrics). It is titled The implications of educational and methodological background for the career success of Nobel Laureates: An investigation of major awards. The second part of the thesis (Chapter 4 and 5) contributes to the empirical research on scientific collaboration patterns. These are joint works with Ali Sina Önder and Benno Torgler and both are published in Scientometrics. Chapter 4 corresponds to the essay Do Nobel laureates change their patterns of collaboration following prize reception?, and Chapter 5 corresponds to the study titled The first cut is the deepest: Repeated interactions of co-authorship and academic productivity in Nobel laureate teams. The two studies exploring the correlation between scholars’ external and internal influence and their relation to public speaking fees are presented in Chapter 6 and 7. These two chapters, titled External influence as an indicator of scholarly importance (published in CESifo Economic Studies) and Do the best scholars attract the highest speaking fees? An exploration of internal and external influence (published in Scientometrics) are the results of collaborative work with Bruno Frey, Jana Gallus, Markus Schaffner, Benno Torgler and Stephen Whyte. Next, I present the study titled The inner quality of an article: Will time tell? which was completed in collaboration with Malka Guillot, Lionel Page and Benno Torgler (Chapter 8, in Scientometrics). Lastly, in Chapter 9, I present the essay Do Great Minds Appear in Batches? co-authored with Benno Torgler, which sets out to explore the emergence of great scientists in the last two centuries (in Scientometrics). In the conclusion chapter, I provide the summary of key findings and limitations of each study. I then conclude by focusing on the future direction of the current work in this field.

Chapter 2 Awards Before and After the Nobel Prize: A Matthew Effect and/or a Ticket to One's Own Funeral?

Chan Ho Fai, Laura Gleeson & Benno Torgler

Research Evaluation (2014), 23(3), 210-220.

Abstract

The primary aim of this descriptive exploration of scientists’ life cycle award patterns is to evaluate whether awards breed further awards and identify researcher experiences after reception of the Nobel Prize. To achieve this goal, we collected data on the number of awards received each year for 50 years before and after Nobel Prize reception by all 1901–2000 Nobel laureates in physics, chemistry, and medicine or physiology. Our results indicate an increasing rate of awards before Nobel reception, reaching the summit precisely in the year of the Nobel Prize. After this pinnacle year, awards drop sharply. This result is confirmed by separate analyses of three different disciplines and by a random-effects negative binomial regression model. Such an effect, however, does not emerge for more recent Nobel laureates (1971– 2000). In addition, Nobelists in medicine or physiology generate more awards shortly before and after prize reception, whereas laureates in chemistry attract more awards as time progresses.

Introduction

Frey (2006: 377) remarks that ‘[i]f an alien were to look at the social life of people here on earth, it would be stunned by the enormous number of awards in the form of orders, medals, decorations, prizes, titles, and other honours. It would be hard pressed to find any area of society in which awards are not used’. Universities and the academic environment in general have developed an extensive system of awards (Frey and Neckermann 2009), based on a long

tradition of recognising scientific achievements through award conferral dating back to the 18th century (Zuckerman 1992). For example, the Copley Medal, the oldest and most prestigious award of Britain’s Royal Society, was first awarded for outstanding achievements in the physical and biological sciences in 1731, since when it has been given to such notables as Charles 2 Darwin and Michael Faraday. The Nobel Prize, however, is considered the ultimate accolade in science (Merton, 1968), and even American director and comedian Woody Allen (despite consistently refusing to attend the Academy Award ceremonies when nominated for his films) admitted that he would show up for a Nobel Prize: ‘A Nobel prize would be different’, Allen observed, ‘apart from everything else . . . it carries an interesting amount of cash’ (Zuckerman 1992: 219). Frey and Osterloh (2010: 871) further note that the ‘incentive system for scholars has to match their main motivation factors. Prizes and titles are better suited for that purpose than citation metrics. Honorary doctorates, different kinds of professorships and fellowships (from assistant to distinguished), membership of scientific academies, and honours such as the Fields Medal or Nobel prizes are great motivation even for those who do not actually win such a prize. The money attached to such rewards is a bonus, but less important than the reputation of the award-giving institution’. Economists have described this reward process as a non-market- based incentive system to produce the public good of knowledge, one that compensates individuals through achievements in job positions in which it is difficult to monitor effort (Stephan 2012).

The aim of this study, therefore, is to descriptively explore the award pattern of Nobel laureates before and after the Nobel Prize using key concepts capable of advancing theoretical understanding of award patterns over time, such as the presence of a Matthew effect and/ or the adverse effects of awards and recognitions. We also address the important issue of whether our results could be driven by an age effect rather than a recognition or award effect by taking the potential for both age and productivity effects into account in the empirical analysis.

2 See http://royalsociety.org/awards/copley-medal/.

Theoretical considerations

Matthew effect in science According to Merton (1968: 159), the potential ‘Matthew effect’ in the academic reward system ‘consists of the accruing of greater increments of recognition for particular scientific contributions to scientists of considerable repute and the withholding of such recognition from scientists who have not yet made their mark. Nobel laureates provide presumptive evidence of the effect, since they testify to its occurrence, not as victims—which might make their testimony suspect—but as unwitting beneficiaries’. Referring to interviews conducted with Nobel laureates, Merton (1968: 2) comments that ‘[t]he world is peculiar in this matter of how it gives credit. It tends to give the credit to [already] famous people’, a sentiment echoed by Nobel laureate Herbert Simon, who remarks that ‘after a while the criterion for getting an honor is to have been awarded a lot of other honors’ (as cited in Klahr 2004: 440). Zuckerman (1996: 237) also stresses that ‘laureates become prime candidates for other honors, since association with the Nobel prize, as we have noted, seems to enhance the prestige of other awards and the standing of the organisations that confer them. Choosing laureates has advantages; those responsible for selecting recipients obviously do not wish to make mistakes and so they protect themselves by giving awards where the Nobel has already committed itself’.

It is Zuckerman (1996), in fact, who makes the important distinction between additional and multiplicative effects when discussing the accumulation of advantages over time. Whereas the additive model suggests that those who begin generating advantages keep benefiting and receiving resources and rewards irrespective of performance (the accumulative advantage hypothesis, see Allison et al. 1982 for a discussion), the multiplicative model stresses their accumulation of more of the factors required to increase achievements, which allows them to move ever farther into the forefront. Recognition can thus be transformed into resources for further work, while good work leads to an increased esteem from colleagues, which attracts more recognition. This cycle leads to a Matthew effect in the distribution of honorific awards: ‘those who already have them are most likely to receive new ones’ (Zuckerman 1996: 63). It could also be, however, that experiencing success increases the taste for further success, motivating scientists to work harder (Stephan 2012).

The literature on the Matthew effect in the area of scientific work and beyond is large (see, e.g., Cole 1970; Allison and Stewart 1974; Goldstone 1979; Allison et al. 1982; Walberg and Tsai 1983; Stanovich 1986; Hunt and Blair 1987; Merton 1988, 1995; Rossiter 1993; Bonitz et al. 1997; Van Looy et al. 2004; Medoff 2006; Strevens 2006; Tol 2009; Bothner et al. 2010, 2011; Petersen et al. 2012; Olry and Haines 2012; Stephan 2012; Wang 2014). A substantial number of such studies focus on citations (see, e.g., Cole 1970 for a powerful early study), which permits evaluation of citation diffusion (i.e., such citation dynamics as how promptly and widely they emerge) and thus the communication path of insights and the network characteristics of knowledge transfer. Receiving a major academic award, however (as opposed to accumulating citations), provides much greater visibility within the scientific community and beyond, particularly in the past when citation data were less accessible. Today, free access to such resources as Google Scholar or Microsoft Academic Search allows individual citation performance to be monitored on an almost daily basis. Basing analysis on citations, however, is problematic in that they can be criticised as measuring usefulness to other scientists rather than quality or contribution to society in general (Seglen 1992). Award reception, although it has the advantage of offering a clearly dichotomous variable (an individual either has or has not received the prize), can also suffer from weakness. That is, unlike the individual choice to cite a paper, which is the prerogative of every scholar, award recipients are often selected by assigned experts. Thus, bestowal of awards is regulated in a particular direction, that is, towards the promotion and recognition of the novelty or magnitude of the knowledge produced.

Admittedly, papers already exist that examine awards and recognitions; Azoulay et al. (2014), for example, by investigating the effect of the positive status shock of appointment as a life scientist investigator at a highly regarded biomedical institution, find evidence of a moderate post- appointment citation boost in articles published before the appointment. Likewise, Chan et al. (2013) assess the consequences of receiving the John Bates Clark Medal, a prestigious award in economics given to a US scholar under 40 years of age, by comparing medal winners to a group of nonrecipient scholars with similar pre-prize publication and citation performance. Their analysis of the differences in both groups’ post-prize research productivity patterns and citation patterns of pre-prize publications seemingly

12 indicates that John Bates Clark medallists publish more and have their pre- prize work cited more often than do the control group. Nevertheless, not only is it very difficult to isolate the exact factors that determine success, but the committee selection process is crucial in any such analysis. That is, if the committee’s choice is driven by unobservable winner characteristics relative to non-winners and these unobservable characteristics are correlated with changes in research productivity, biases may emerge.

In general, we are unaware of any other study that examines a particular time window before and after the Nobel Prize to assess how the number of awards changes over time. One important caveat, however, is that we cannot measure a Matthew effect in the classical sense, although as the already quoted material strongly suggests, a Matthew effect by which ‘success breeds success’ and ‘the rich get richer’ could account for an increase in awards over time. Nor can we conclude from the available data that such a trend results from a misallocation of credit, which is the key problem in the Matthew effect.

Such misallocation of credit is exemplified by Merton’s (1968) case of multiple discoveries, which offers a ‘cleaner’ framework than the one we provide (for an excellent discussion, see Cole 1970). If, for example, two scientists independently make the same discovery, under the Matthew effect, the more eminent of the two will receive the greater credit. Such misallocation of credit can have negative externalities for science by unjustly influencing scientists’ career. Hence, based on Harriet Zuckerman’s interviews with Nobel laureates, Merton (1968) concludes that laureates themselves perceive the Matthew effect as a problem (i.e., a simple allocation of credit is not achieved), one that creates a positive and negative multiplier (i.e., by unjustifiable victimising unknown scientists while unjustifiably benefiting famous ones). Accordingly, he stresses that ‘[w]hen the Matthew effect is thus transformed into an idol of authority, it violates the norm of universalism embodied in the institution of science and curbs the advancement of knowledge’ (Merton 1968: 7). Cole (1970), however, points out that the ideal test for multiple discoveries is problematic because they are difficult to identify.

Possible adverse effects of awards As illustrated by the following recollections from the period shortly before the announcement, directly after the announcement, and during the trip to Stockholm, receiving a Nobel Prize is a wonderful experience for a scientist. Samuelson (2004: 62), for example, recalls the following: ‘Someone asked me whether I enjoyed getting the Nobel Prize. I thought before answering. “Yes”, I replied, “few things in life bring undiluted pleasure, but this one actually did”. The honor was a pleasant surprise and came early, but not so early as to worry even me. Friends whose opinions I valued were pleased. If here were contrary opinions, I was too obtuse to be aware of them. My family enjoyed the Stockholm hoopla. Some colleagues in science have looked back with pain at the public interviews and turmoil that took them out of their laboratories. I bore up well and discovered that it takes only a few days of dependence upon one’s own chauffeur to develop an addiction.’ James Watson (2007: 177) particularly remembers receiving some 80 congratulatory telegrams in the first 2 days after the Nobel Prize announcement, among them the following words from Richard Feynman: ‘AND THERE HE MET THE BEAUTIFUL PRINCESS AND THEY LIVED HAPPILY EVER AFTER’. Herbert Simon (1996: 319, 323) describes the post-prize experience as follows: ‘More than 10 years after the award of the Nobel Prize, people, on meeting me, still congratulate me, as though they have been remiss in holding silence for a decade . . . . October 15, 1978, was a Sunday. A week or two earlier, my name had been printed in a Swedish business magazine as one of the short-list candidates for the 1978 Nobel Prize, and I opened my newspaper each morning that week to search for the verdict there. On Sunday afternoon, I received a phone call from my friend and former student Sven-Ivan Sundqvist, who had arranged my Stockholm visit in 1969. Sven reported that he had met a member of the Nobel Committee on the street that day who told him he would not be disappointed by that year’s award. After considering what that might mean, he decided that I would be the winner and called to alert me. Needless to say, I found myself a bit tense and exhilarated during the rest of the afternoon, and made plans to arise early the next morning in case a phone call came from Stockholm after the academy meeting that was to end there about noon, Stockholm time. When the call came, at 6:00 A.M. on Monday, I was already up and dressed.’

A famous quote by Tom Eliot, Nobel laureate in literature, however, indicates that he saw the recognition more as an epitaph than an award: ‘The Nobel Prize is a ticket to one’s own funeral. No one has ever done anything after he got it’ (Meyers 2007: 221). This latter would suggest Faustian aspiration that has come to an end: ‘If ever I to the moment shall say: Beautiful moment, do not pass away! Then you may forge your chains to bind me.’ After becoming a Nobel laureate, therefore, a scientist may be less concerned about subsequent honours or awards (Zuckerman 1996), although, as indicated by economics laureate Paul Samuelson, recognition can be a strong driving force (2004: 60):

Scientists are as avaricious and competitive as Smithian businessmen. The coin they seek is not apples, nuts, and yachts; nor is it the coin itself, or power as that term is ordinarily used. Scholars seek fame. The fame they see . . . is fame with their peers—the other scientists whom they respect and whose respect they strive for. The sociologist Robert K. Merton has documented what I call this dirty little secret in his book The Sociology of Science. I am no exception. Abraham Lincoln’s law partner and biographer William Herndon observed that there was always a little clock of ambition ticking in the bosom of honest and whimsical Abe. No celebrity as a Newsweek columnist, no millions of clever-begotten speculative gains, no power as the Svengali or Rasputin to the prince and president could count as a pennyweight in my balance of worth against the prospect of recognition for having contributed to the empire of science.

Merton (1973: 341) also cites Selye’s comments on such recognition: ‘Why is everybody so anxious to deny that he works for recognition? ... All the scientists I know sufficiently well to judge (and I include myself in this group) are extremely anxious to have their work recognised and approved by others. Is it not below the dignity of an objective scientific mind to permit such a distortion of his true motives? Besides, what is there to be ashamed of?’

Another aspect of the post-prize experience, according to Zuckerman (1996), is that laureates are more hesitant to publish work that might be judged as weak, seemingly responding to a raised personal standard and a perceived standard expected from others. She uses the following remark by a physicist to exemplify such reluctance: ‘After you’ve done something good and received such high recognition for it, it’s hard to publish anything without feeling it’s below the stature you’ve gained. It becomes very hard to do anything that you might call pedestrian, and a good many people just quit. At the present time, it’s difficult for me to keep going because of all of this extraneous honor’ (Zuckerman 1996: 229). Merton (1968: 2) also points to the challenges faced after a major award like the Nobel Prize: ‘The scientist’s peers and other associates regard each of his scientific achievements as only the prelude to new and greater achievements. Such social pressures do not often permit those who have climbed the rugged mountains of scientific achievement to remain content … More and more is expected of them, and this creates its own measure of motivation and stress.’

What about age? Given that we are exploring recognition over time, it is worth discussing the possible importance of the age factor. Unfortunately, the literature on the relation between age and recognition remains underdeveloped, while the literature on age and productivity or creativity is so extensive—dating back even before Lehman’s (1953) seminal Age and Achievement (e.g., Adams 1946)—that it cannot be adequately addressed in this brief discussion. The most relevant work for our present analysis, however, includes Simonton’s (1988a,b, 1992) useful overviews of studies on age and outstanding achievement and Stephan and Levin’s (1993) brilliant article on age viewed from such different perspectives as creativity or the willingness to do science. Levin and Stephan (1991), Jones (2010), and Jones and Weinberg (2011) also offer careful empirical analyses on age and productivity, age and inventions, and age and scientific creativity, respectively. Such literature is relevant to our context because, if awards and recognitions are indeed driven by productivity and/or creativity, it is important to understand the relation between these two variables and age. That is, awards may lag behind productivity, but there is also some indication that recognition is conferred relatively rapidly on Nobel laureates after their peak scientific achievement (Chan and Torgler 2013).

The possibility that our analytical results could be driven by an age effect stems from the reality that motivation and actual work effort may decrease over time, the taste of success may change, and ‘sacred sparks’ may decrease (Stephan and Levin 1993). Hence, analytical outcomes could be driven not only by past award history (success or failure in obtaining awards) but also by other factors such as a decline in mental processes, in generating interesting questions, or in energy level, combined with an increase in other activities (e.g., administrative work instead of research). Controlling for age may help take into account some of these effects, especially those highly correlated with age.

In general, the research process is facilitated by focusing only on high achievers like Nobel laureates not simply because the productivity in science is highly skewed but also because Nobel laureates or highly successful researchers seem to have (relatively) more stable life cycle performance patterns throughout their careers, together with delayed peak(s) (see, e.g., Cole 1979, Simonton 1988a, Torgler and Piatti 2013). There has in fact been substantial discussion about the functional form of the age curve, which in the tables presented in our study is assumed to be a second-order polynomial approximation for all the fields explored, that is, one that increases at a decreasing rate with only one peak. This latter, although imperfect, follows Simonton (1988a), who concludes in his overview that ‘a single-peak function still provides the most stable summary of the observed data’ (p. 253).

Methods

Data collection Drawing on the very detailed information in Kurian’s (2002) Nobel Scientists: A Biographical Encyclopedia, we collected data on all 1901–2000 Nobel laureates in physics, chemistry, and medicine or physiology (N=466), looking at the number of awards received each year for 45 years before and 48 years after prize reception (see Appendix Table 2.3, for the top 20 awards received). For Albert Einstein, for example, the encyclopedia lists the 1921 Nobel Prize in Physics awarded ‘[f]or his services to Theoretical Physics, and especially for his discovery of the law of the photoelectric effect’ (Kurian 2002: 141), as well as the Barnard Medal, Columbia University (1920); Copley Medal, Royal Society (1925); Gold Medal, Royal Astronomical Society (1926); Max

Planck Medal (1929); and Franklin Medal, Franklin Institute (1935). It does not, however, include his honorary doctorates from the University of Rostock (1919), Princeton University (1921), University of Madrid (1923), ETH (Eidgenö ssische Technische Hochschule), Zurich (1930), Oxford University (1931), and Harvard University (1935).

An alternative approach to compiling study data would have been to directly collect detailed curriculum vitae (CVs); however, not only is it difficult to obtain such information over an almost 100-year time span, but awards listings tend to be inconsistent. In particular, not all scientists list all their awards on their CVs. In fact, Zuckerman (1996: 238) reports that after receiving the prize, around a fourth of laureates trim their listings in biographical dictionaries (e.g., American Men and Women of Science). For instance, Linus Pauling omitted his multiple honorary degrees and some local prizes, while George von Békésy and Joshua Lederberg did not even list their Nobel prizes.

Statistical analysis The first phase of our statistical analysis is a descriptive exploration of the number of major awards received by the Nobelists before and after the Nobel Prize, followed by a numerical analysis of the relative share of awards among the different disciplines. Because the number of Nobel Prize winners can vary between fields from year to year, we explore the number of awards per number of Nobel Prize winners within a field. To measure the relative proportion, we construct a filter using a 5-year moving average window with two lagged years and two leading years plus the current year. We use the elements on the fifth row of Pascal’s triangle as weights; being the numerator 1 and the sum of the elements as the denominator, that is, t × + -2 16 4 6 4 1 t × + t × + t × + t × . Compared with the uniformly distributed -1 16 16 +1 16 +2 16 weights, this smoothing process provides a more accurate reflection of actual values, since those closer to the actual year are weighted relatively heavier. In addition, for laureates who won the Nobel Prize twice, including Maria Curie (1903 in physics and 1911 in chemistry), John Bardeen (1956 and 1972 in physics), and Frederick Sanger (1958 and 1980 in chemistry), only the first

Prize is used as the cut-off.3 Based on this descriptive analysis, we then conduct a multivariate analysis that estimates the time effect before and after the Nobel Prize and the differences between fields. Specifically, we model the award count of Nobel laureates using a random-effects negative binomial regression model that takes into account the individual heterogeneity of the Nobel recipients. Unlike the Poisson regression model, this model is designed to explicitly handle any over-dispersion observed in our data. It should also be noted that yearly individual observations are dropped once a Nobel laureate passes away. For example, for Ferdinand Frederick Henri Moissan, who received the Nobel Prize in 1906 and died in 1907, only one post-prize year is recorded.

Results

The initial results of our descriptive analysis support the conjecture that success breeds success only up to the point of attaining the Nobel Prize, the highest supreme symbol of accomplishment in science. That is, Figure 2.1 shows an increasing rate of awards before the Nobel Prize, reaching the summit precisely in the year of the Nobel Prize, after which awards drop off sharply. This substantial post-prize decrease does indeed point to ‘negative externalities’ from Nobel Prize reception and the disappearance of any potential Matthew effect. The comparison of relative inter-field differences in awards share—number of awards in a particular year divided by the number of laureates in that field still alive at the year of investigation—again shows that the Nobel year is the peak year for all fields studied (see Figure 2.2). It is also interesting to note that in the periods just before and after the Nobel Prize, recipients from physiology and medicine are generating relatively more awards than those in the other two fields, a trend that changes at a later stage with an increase in awards for chemistry. This latter result, however, is driven by researchers who received the Nobel Prize relatively early in their careers or who lived for many years after prize reception.

3 Linus Pauling (1954 Nobel Prize in chemistry) received the Nobel Peace Prize in 1962; however, the Nobel Peace Prize is not regarded as academic award.

Figure 2.1 Number of major awards before and after the Nobel Prize

Figure 2.2 Share of major awards within different fields (5-year moving average)

The first set of regression results, listed in Table 2.1, support the shape of the descriptive plots. Specification (1) refers to 5 years before and after the Nobel Prize, with the Nobel Prize year as the reference group. As the

20 marginal effects clearly show, the decrease in awards after the Nobel Prize is larger than the increase in awards before the Nobel Prize. Compared with the Nobel Prize year, ceteris paribus, in post-prize Year 4, a Nobel recipient receives 0.326 fewer awards, while in pre-prize Year 5, the difference is only 0.156 awards. We also note that Nobel laureates in physiology or medicine generate more other major awards than those in physics and chemistry. In recognition of the possible age effect, specification (2) includes a control for age. The results, however, change little, confirming the general tendency towards a non-linear structure (inverted U-shape function) in the age– performance relation. Specification (3) then focuses only on the period 6 years and more after the Nobel Prize, a period in which, surprisingly, only chemistry laureates surpass prize recipients in physiology or medicine, albeit with a coefficient that is not statistically significant. This lack of any significant inter-field differences in the career of a Nobel laureate up to 6 years before prize reception is confirmed in specification (4), which identifies no statistically significant differences between physics and chemistry laureates.

Table 2.1 Results of random effects negative binomial regression models

Dependent Variable: Five years Five years Six years Beginning of Number of other major before and before and before and the career awards after Nobel after Nobel after Nobel until six Prize Prize Prize years before the Nobel Prize (1) (2) (3) (4) Awards Year 5 before -0.515*** -0.528*** the Nobel Prize (-3.87) (-3.91) -0.156 -0.159 Awards Year 4 before -0.204* -0.218* the Nobel Prize (-1.68) (-1.77) -0.072 -0.076 Awards Year 3 before -0.358*** -0.37*** the Nobel Prize (-2.83) (-2.91) -0.117 -0.12 Awards Year 2 before -0.347*** -0.355*** the Nobel Prize (-2.75) (-2.81) -0.114 -0.116 Awards Year 1 before -0.027 -0.031 the Nobel Prize (-0.23) (-0.27) -0.010 -0.012 Awards in Year of the Ref. Ref. Nobel Prize Awards Year 1 after the -0.444*** -0.433*** Nobel Prize (-3.42) (-3.34) -0.139 -0.137 Awards Year 2 after the -0.967*** -0.954*** Nobel Prize (-6.35) (-6.25) -0.241 -0.239 Awards Year 3 after the -1.235*** -1.217*** Nobel Prize (-7.26) (-7.15) -0.275 -0.274 Awards Year 4 after the -1.841*** -1.823*** Nobel Prize (-8.41) (-8.29) -0.326 -0.326 Awards Year 5 after the -1.542*** -1.507*** Nobel Prize (-7.83) (-7.63) -0.305 -0.303 Chemistry -0.462*** -0.483*** 0.13 -0.088 (-3.66) (-3.82) (0.94) (-0.71) -0.110 -0.115 0.013 -0.008 Physics -0.433*** -0.433*** -0.135 -0.173 (-3.64) (-3.57) (-0.99) (-1.40) -0.105 -0.106 -0.012 -0.015 Physiology or Medicine Ref. Ref. Ref. Ref.

Female 0.498 0.564* 0.006 -0.082 (1.58) (1.79) (0.01) (-0.23) 0.115 0.131 0.001 -0.007 Age 0.076** 0.071** 0.128*** (2.47) (2.3) (5.94) 0.018 0.007 0.011 Age^2 -0.001*** -0.001*** -0.001*** (-2.68) (-3.22) (-4.78) -0.000 -0.000 -0.000 N 5033 5033 8031 11360 Prob.>χ2 0.000 0.000 0.000 0.000 Notes: Marginal effects in italics, z-statistics in parentheses; *, **, and *** represent statistical significance at the 10%, 5%, and 1% levels, respectively.

The above observations lead us to wonder about productivity over time before and after the Nobel Prize. To address this question, we first conduct a descriptive analysis based on a quality-adjusted proxy that controls for the impact factor of a journal. To measure productivity we collected journal articles of the 1970–2000 Nobel laureates (N=198) from Scopus4, a bibliometric database for peer-reviewed articles. The publication dates range from 1927 to 2013 with a total of 36,344 articles5. We excluded conference papers, reviews, notes, surveys, letters, and errata and we considered only those publications written in English. Moreover, we manually excluded all Nobel Lectures which were published as journal articles. Data on the journal impact factors were obtained from Journal Citation Reports (JCR) and were merged with the publication data resulting in 31,437 articles, providing us with full data on 86.5% of publications (as not all are indexed by JCR6). Based on JCR 2012 data, we take only recently published impact factor values (average 5-year journal impact factor for 2007–2011). Obviously, journal impact factors change over time—some journals are relatively new, while others disappear—however, even recent data could be a good proxy for relative journal quality over time (particularly for the leading publications in which Nobel laureates frequently publish). We therefore multiply each single Nobelist’s publication recorded in Scopus by the corresponding journal impact factor and sum up the publications within a year by taking the average of all the Nobelists’ output within that year (since the Nobel Prize). Using the same 5-year moving average window as before, Figure 2.3 reveals the publication pattern for Nobelists in different fields. All three fields experience an inverted U-shape function in their publication tracks, affirming the findings from the previous age-productivity literature (e.g., Levin and Stephan 1991, Jones 2010, and Jones and Weinberg 2011).

4 http://www.scopus.com/ 5 An additional measure of productivity would be to incorporate the number of patents a scientist holds. For example, Guglielmo Marconi, a pioneer in long-distance radio transmission and the 1909 Nobel Prize winner in Physics, has over 200 patents in France, the UK, and USA. For studies using patents as productivity measure, see Jones (2009), Jones (2010), and Schettino et al. (2013). 6 Impact factors for some older journals were not available because of splits and merges, for example, Physical Review was divided into four sub-journals after 1970 (with 622 articles published by the 1970–2000 Nobel laureates), Journal of Physical Chemistry was split into Journal of Physical Chemistry A and B in 1997 (373 articles before the split). Journal of the Chemical Society was split into three sub-journals in 1966 (before 1966, 128 articles).

Figure 2.3 Publication performance pre and post Nobel Prize

We observe that before receiving the Nobel Prize, Nobel laureates in physiology or medicine publish more intensively compared with Nobelists in chemistry and physics (in this order). For example, the average weighted publication counts from 5 years before (t_5) to 1 year before obtaining the Nobel Prize (t_1) are 59.9 for physiology or medicine, 35.7 for chemistry, and 12.1 for physics7. While publication counts in physiology or medicine and physics drop after reception, laureates in the field of chemistry continue publishing more articles until 5 years after obtaining the Nobel Prize, reaching the summit of 41.8 weighted publications.

The patterns revealed by Figure 2.3 motivated us to include an additional control for (adjusted) productivity (see Table 2.2), one first applied in the form of a control for a Nobelist’s yearly productivity [number of publications, see specifications (1) to (3)]. For convenience, we applied this control only to 1970–2001 laureates, collecting all their publications available on Scopus. In this initial analysis, we use three measures of productivity: the number of yearly publications (specification 1), the number of yearly

7 Reported values are the average of the actual publication count. Nobel laureates in Physiology or Medicine publish on average 6.79 articles in journals with an average impact factor of 8.02 annually during this period.

24 publications divided by the square-root of the number of authors (specification 2), and the number of yearly publications adjusted for the previously discussed quality proxy (specification 3). Specification 2, which employs what is also known as Price’s ‘square root law’ (see, e.g., Glänzel and Schubert 1985, Nicholls 1988), allows us to take into account non- linearity in the weighting process.

The results for these productivity measures indicate that the decrease in the number of awards after the Nobel Prize is very robust. Interestingly, however, the positive increase in pre-Nobel awards observed earlier now disappears. Hence, to gauge whether this puzzling result was driven by the addition of the productivity proxies (all statistically significant with a positive sign), we developed specification (4), which omits these three indexes. In this case, we find that the results for the pre-prize year are not significantly different from those for the Nobel Prize year, which may imply a cohort effect. To evaluate this conjecture, we run an additional regression (specification 5) for the remaining set of laureates (1901–1969). Interestingly, the results for this latter are comparable with the Table 2.1 outcomes suggesting that earlier success matters. Testing this observation by going back a mere 30 years (1940–1969) produces similar results.

We therefore ponder whether it is viable to conclude that for more recent cohorts of Nobel laureates, there is no sign of success breeding success or a Matthew effect. Could it not be rather that competition (i.e., publish or perish!) has become more serious? The overflow of publications produced per year is massive, increasing also over time as newly emerged journals ‘battle for attention’ (Torgler and Piatti 2013). Yet finding high quality contributions requires time and energy that are in short supply, so as complexity increases, journal editors rely more on simple heuristics in their decision process (Stadelmann and Torgler 2013). One could therefore argue that awards serve an important signalling function (Frey and Gallus 2014). Such an argument, however, is more convincing in the citation context, in which a scholar searching for a particular topic is more likely to read and possibly cite the work of an eminent researcher (e.g., Nobel laureate). Yet Chan et al. (2013) find a larger citation gap between their control group and the 1947–1975 John Bates Clark medallists than between this same control group and 1977–2001 medallists. Specifically, early Clark medallists have an 81% higher citation rate than the control group versus 76.9% for the later

25 medal winners. Moreover, in any examination of award accumulation over time, both supply and demand effects should play a role. That is, as the number of awards has grown drastically over time (Stephan 2012), award providers may have begun to think more about how to guarantee distinction. In particular, despite the tendency for superstars to emerge in highly competitive markets (Frank and Cook 1996), if the research market has indeed become more competitive, then the supply of potential candidates with comparable skills has become larger, reducing the risk of seeming to choose the wrong person. Hence, giving the same award again to an already successful researcher (in terms of awards) may be seen as insufficiently distinct. Such conjectures, however, are mere speculations that need to be substantiated through more research on the motivation of awards suppliers. More exploration is also needed on different cohorts over time.

Table 2.2 Awards, productivity, and cohorts

Dependent Nobel Prize Nobel Prize Nobel Prize Nobel Nobel Variable: Laureates Laureates Laureates Prize Prize Number of 1970-2000 1970-2000 1970-2000 Laureates Laureates other Major 1970-2000 1901-1969 Awards Awards Year -0.177* -0.188 -0.151 -0.071 -0.978*** 5 before the (-0.94) (-1.00) (-0.81) (-0.38) (-4.93) Nobel Prize -0.061 -0.065 -0.052 -0.025 -0.25 Awards Year 0.116 0.107 0.141 0.209 -0.622*** 4 before the (0.66) (0.62) (0.81) (1.21) (-3.54) Nobel Prize 0.046 0.043 0.056 0.086 -0.186 Awards Year 0.046* 0.039 0.069 0.133 -0.887*** 3 before the (0.26) (0.22) (0.39) (0.75) (-4.68) Nobel Prize 0.018 0.015 0.026 0.052 -0.236 Awards Year -0.11* -0.118 -0.111 -0.066 -0.597*** 2 before the (-0.60) (-0.64) (-0.6) (-0.35) (-3.46) Nobel Prize -0.039 -0.042 -0.039 -0.023 -0.18 Awards Year 0.286 0.285* 0.299* 0.311* -0.336** 1 before the (1.70) (1.70) (1.77) (1.83) (-2.12) Nobel Prize 0.125 0.124 0.129 0.134 -0.115 Awards Year Ref. Ref. Ref. Ref. Ref. of the Nobel Prize Awards Year -0.291* -0.294 -0.281 -0.311 -0.541*** 1 after the (-1.49) (-1.5) (-1.43) (-1.57) (-3.18) Nobel Prize -0.095 -0.096 -0.091 -0.098 -0.168 Awards Year -0.958*** -0.962*** -0.964*** -0.987*** -0.96*** 2 after the (-3.97) (-3.99) (-3.99) (-4.07) (-4.94) Nobel Prize -0.232 -0.233 -0.229 -0.23 -0.248

Awards Year -1.021*** -1.027*** -1.008*** -1.052*** -1.385*** 3 after the (-4.05) (-4.07) (-4) (-4.16) (-6.04) Nobel Prize -0.24 -0.242 -0.235 -0.239 -0.301 Awards Year -1.873*** -1.877*** -1.863*** -1.914*** -1.801*** 4 after the (-5.22) (-5.24) (-5.19) (-5.34) (-6.57) Nobel Prize -0.318 -0.319 -0.312 -0.313 -0.335 Awards Year -1.472*** -1.468*** -1.474*** -1.568*** -1.543*** 5 after the (-4.6) (-4.6) (-4.59) (-4.86) (-6.24) Nobel Prize -0.29 -0.29 -0.285 -0.291 -0.316 Chemistry -0.392** -0.406** -0.337* -0.398** -0.578*** (-2.23) (-2.33) (-1.91) (-2.19) (-3.34) -0.117 -0.119 -0.099 -0.123 -0.11 Physics -0.411** -0.377** -0.359** -0.509*** -0.425*** (-2.37) (-2.18) (-2.02) (-2.96) (-2.63) -0.122 -0.112 -0.105 -0.149 -0.087 Physiology or Ref. Ref. Ref. Ref. Ref. Medicine Female 0.746 0.77* 0.763* 0.437 0.336 (1.87) (1.95) (1.95) (1.08) (0.68) 0.215 0.221 0.218 0.125 0.064 Age 0.057* 0.055 0.065 (1.15) (1.1) (1.32) 0.017 0.016 0.019 Age^2 -0.001* -0.001 -0.001* (-1.55) (-1.48) (-1.69) -0.000 -0.000 -0.000 Number of 0.023*** Publication (2.67) 0.007 # Publication 0.055*** / SQRT(# (3.21) Authors) 0.016 # Publication 0.003*** * 5-Year (2.73) Journal Impact Factor 0.001 N 2121 2121 2121 2121 2912 Prob.>χ2 0.000 0.000 0.000 0.000 0.000 Notes: Marginal effects in italics, z-statistics in parentheses; *, **, *** represent statistical significance at the 10%, 5%, and 1% levels, respectively.

Discussion

Overall, our results suggest that a constant increase in awards—which could indicate a Matthew effect in the sense of ‘success breeding success’—only holds up to the point of receiving the Nobel Prize and seems to depend on the cohort analysed. Moreover, we observe no constant and significant increase

27 in the 2 years before the Nobel Prize for more recent Nobel laureates; in fact, for laureates, the number of awards decreases substantially after prize reception. This result is very robust, even among different cohorts. Our findings, therefore, seem not to support Zuckerman’s (1996) contention that laureates become prime candidates for honours because the Nobel Prize (1) increases their standing and therefore also the prestige of the other awards they receive and (2) reduces the risk of selection committee mistakes in awardee choice.

Nevertheless, we admit that our results could be driven by our focus on major awards only, meaning that outcomes could be changed by the inclusion of such additional accolades as honorary doctorates. Future research, therefore, might consider differentiating between major and minor awards. Other questions worthy of exploration are what Nobel laureates do once they receive the Nobel Prize, what explains the difference between those who receive fewer awards after the Nobel Prize and those who are more successful, and whether those who receive fewer awards are more engaged in other activities such as setting up new labs, teaching, giving Nobel lectures, or simply retiring.

The shape of the curve identified in our analysis may also be driven by how organizations award prizes. For example, one could argue that the academy has an incentive to avoid premature judgments by awarding the prize long after researchers have already earned academic fame. Yet Chan and Torgler (2013) provide clear evidence that recognition is actually conferred relatively rapidly in academia. Moreover, once a scientist has climbed to the summit of scientific achievements, other award providers may have a lower incentive to offer this personality an additional accolade that could only live a shadowy existence next to the Nobel Prize. In fact, Zuckerman (1996: 34) suggests that ‘some organizations actively resist the tendency to have their evaluations in effect pre-empted by the academies in Stockholm. Thus, a member of a university committee on honorary degrees remarked to me that his colleagues refused to follow along after the Nobel’.

Appendix

Table 2.3 Top 20 major awards obtained by Nobel laureates (1901-2000) by field

Total Physiology Awards Chemistry Physics number of or Medicine awardees National Orders 30 41 45 116 National Medal of Science 14 21 20 55 Lasker Award 6 1 44 51 Copley Medal 11 15 18 44 Royal Medal 12 14 13 39 Les Prix Canada Gairdner 7 0 28 35 Awards Franklin Medal 12 21 1 34 Davy Medal 26 2 0 28 Louisa Gross Horwitz 6 0 21 27 Prize Hughes Medal 2 25 0 27 Faraday Medal 15 11 0 26 Elliott Cresson Medal 5 16 0 21 Cameron Prize 0 0 20 20 Wolf Prize 6 6 6 18 John Scott Medal 6 4 7 17 Max Planck Medal 1 14 1 16 Rumford Medal 5 11 0 16 Passano Award 2 0 13 15 Humboldt Prize 5 8 0 13 Research Corporation for Science Advancement 2 7 3 12 (RCSA) Other Awards 515 527 752 1,794 Notes: Values indicate the number of Nobel laureates who received the listed awards; some awards were conferred more than once to the same person and were awarded to scholars from another discipline who had made contributions to a specific field. For example, the Franklin Medal, a scholarly and engineering award, was awarded to Stanley Cohen (1986 Nobel Prize in Physiology or Medicine) in the subject area Life Science in 1987. Marie and Pierre Curie (1903 Nobel Prize winners in Physics) were conferred the Davy Medal in 1903 by the Royal Society of London which recognises ‘outstandingly important recent discovery in any branch of chemistry’.

Chapter 3 The Implications of Educational and Methodological Background for the Career Success of Nobel Laureates: An Investigation of Major Awards

Chan Ho Fai & Benno Torgler

Scientometrics (2015), 102(1), 847-863.

Abstract

Nobel laureates have achieved the highest recognition in academia, reaching the boundaries of human knowledge and understanding. Owing to past research, we have a good understanding of the career patterns behind their performance. Yet, we have only limited understanding of the factors driving their recognition with respect to major institutionalised scientific honours. We therefore look at the award life cycle achievements of the 1901–2000 Nobel laureates in physics, chemistry, and physiology or medicine. The results show that Nobelists with a theoretical orientation achieved more awards than laureates with an empirical orientation. Moreover, it seems their educational background shapes their future recognition. Researchers educated in Great Britain and the US tend to attract more awards than other Nobelists, although there are career pattern differences. Among those, laureates educated at Cambridge or Harvard are more successful in Chemistry, those from Columbia and Cambridge excel in Physics, while Columbia educated laureates dominate in Physiology or Medicine.

Introduction

Recognition is a key driving force of human nature. Scientists are not exempt from this desire; in fact the academic environment relies on an extensive system of awards (Frey 2007, Chan et al. 2014a). As Zuckerman (1996, p.

209) suggests, the prime ethos of science should be to derive gratification from conducting research and contributing new knowledge. However, research by Merton (1973) highlights the frequent conflicts over priority, indicating that even most eminent scientists care about recognition. As an institutional instrument in academia, awards remind scientists to advance knowledge, while providing them with a chance to find “those happy circumstances in which self-interest and moral obligation coincide and fuse” (Merton 1973, p. 293). Recognition can be an end in itself, but it can also be an instrument through which engagement in scientific activities and achievements is maintained, particularly in an environment where monitoring is difficult (Stephan 2012) or where reputation requires constant replenishment (Zuckerman 1996). Scientists care about their peers, searching for their approval and praise as indicated by Charles Darwin’s statement [cited in Merton (1973, p. 293)]: “My love of natural science … has been much aided by the ambition to be esteemed by my fellow naturalists”.

A growing institutionalization and professionalization of science in the last century has increased discussion regarding how a researcher’s educational background shapes his/her future success. Previous studies have found that successful scientists are more likely to be educated in top institutions. For example, Zuckerman (1996) reports evidence on doctoral origins of US Nobel laureates (1901–1972). A few large entities drive most of the action8: five universities are responsible for more than 55 % of all the US laureates with Harvard being the most successful Ph.D. conferring institution (16.2 %). Thus, it seems that the early academic beginning has an influence on how an academic career develops, with the environment in which one is educated and trained shaping further success. In addition, we may have a selection effect as talented and capable students are attracted to these places. As Zuckerman (1996) points out: “[T]he clumping of future members of the scientific ultra-elite in elite institutions begins early in the selective educational process” (p. 83). “…few elite universities had a distinct advantage in getting to observe and assess these talented young scientists during an important formative phase of their careers” (p. 151). She also shows that more than half of the laureates had previously worked either as students,

8 Such a skewed distribution is common in all kind of academic environments and settings with respect to publication and citation success (see, e.g., Hirsch et al. 1984, Hogan 1986, Cox and Chung 1991, Torgler and Piatti 2013 in the area of economics).

32 post-doctorates, or junior collaborators under Nobel laureates. Hardly any had studied under an unproductive scientist. She refers to Samuelson’s acceptance speech at Stockholm: “I can tell you how to get a Nobel Prize. One condition is to have great teachers” (p. 106) and to Hans Krebs who argued that he owed the Nobel Prize to the “circumstance that I had an outstanding teacher at the critical stage in my scientific career” (pp. 124–125). Zuckerman also emphasizes that “the lines of elite apprentices to elite masters who had themselves been elite apprentices, and so on indefinitely, often reach far back into the history of science, long before 1900, when Nobel’s will inaugurated what now amounts to the International Academy of Sciences” (p. 105). Leading researchers transfer to their youngsters the norms and values for significant research and how to cope with chosen problems (Merton 1968). Interview responses reported by Zuckerman (1996) show that Nobelists benefited from seeing their masters operate, perform, think; or more generally observing their method of work, their standards, and the research culture.

Galton (1874) devoted the last chapter of his famous book English Men of Science to the importance of education. He noted that one-third who replied to his inquiry were educated at Oxford or Cambridge, another third at Scottish, Irish, or London universities and the rest did not declare a university affiliation. He discusses some of the positive educational experiences that respondents had such as “[s]ufficient groundwork on many subjects to avoid error” or “well-balanced education (including chemistry, botany, logic and political economy” (p. 253). He summarizes the results he obtained with the following: “To teach a few congenial and useful things very thoroughly, to encourage curiosity concerning as wide a range of subjects as possible, and not to overteach” (p. 256). Besides stressing the usefulness of applying the subjects of “mathematics”, “logic”, “observation, theory and experiment”, “accurate drawing of objects” connected to the pursued science, and “mechanical manipulation” (p. 256) he emphasizes the importance of self- studying: “This seems sufficient, because my returns show that men of science are not made by much teaching, but rather by awakening their interests, encouraging their pursuits when at home, and leaving them to teach themselves continuously throughout life” (p. 257).

A key distinction in academia is made between researchers who are theorists and those who are empiricists. It is therefore worth exploring whether such a specialization impacts future recognition. There is evidence

in economics that empirically oriented research is cited more often that theoretical research (Johnston et al. 2013). However, with respect to awards and recognition, a tension emerges when discoveries are made: Who should get the recognition? Should it be the theorist who proposed the idea (or the theoretical framework), or should it be the empiricist who provides the 9 evidence? One possible criticism of empirics is that they are useless without the theoretical foundation. It is like a setting out for a walk without a compass and a map or no clue where to go. For example, Kuhn (1977) stresses: “To discover quantitative regularity one must normally know what regularity one 10 is seeking and one’s instruments must be designed accordingly” (p. 219). Wilson (1998) argues: “Nothing in science—nothing in life, for that matter— makes sense without theory. It is our nature to put all knowledge into context in order to tell a story, and to re-create the world by this means” (p. 56). On the other hand, Nobel laureate Simon (2008) points out: “I think we have been sold a bill of goods with the argument that we must always have a clear-cut theory before we can do empirical work…It is argued that if you don’t have 11 a hypothesis you are just counting bricks. But is that a bad thing? If you look down the list of outstanding discoveries in the physical sciences or the biological sciences—look at Nobel awards in those fields—you will note that a considerable number of the prizes are given to people who had the good fortune to experience a surprise” (p. 22). Norrby (2010, pp. 47–58) provides a detailed discussion of serendipitous events that resulted in Nobel Prizes. Creativity enhanced by obsession and inspiration sparked by accidental events of (what Pasteur referred to as) an esprit preparé (prepared mind) led to outstanding achievements (p. 47). Experimental research in various disciplines offers a good example for the proposition that the road to scientific knowledge is not a one-way street from scientific law to scientific measurement (as Kuhn (1977) emphasizes when discussing the development

9 Kolbert (2007) nicely points out this problem in an article in The New Yorker, discussing particle physics and the work environment around CERN’s Large Hadron Collider. 10 To better clarify his point: qualitative research which can be theoretical but also empirical is the prerequisite to a successful quantification (p. 213). 11 “… the methodological directive, “Go ye forth and measure,” may well prove only an invitation to waste time. If doubts about this point remain, they should be quickly resolved by a brief review of the role played by quantitative techniques in the emergence of the various physical sciences” Kuhn (1977, p. 213).

34 of physical science); but rather it is a busy intersection where speaking to theorists and searching for facts is common practice.

In this study we analyse the award life cycle achievement of the 1901–2000 Nobel laureates in physics, chemistry, and physiology or medicine. Previous research has examined the relationship between age and performance/creativity among Nobel laureates (Stephan and Levin 1993, Jones and Weinberg 2011). However, to our knowledge, there is no existing detailed empirical analysis on academic career and award dynamic. In a previous study (Chan et al. 2014b) we explore the number of awards received by a Nobelist before and after obtaining the Nobel Prize. The results indicate a strong increasing rate of awards before the Nobel Prize, reaching the summit in the year of the Nobel Prize followed by a very strong decrease after the Nobel Prize. The current study goes beyond our prior work by using a lifecycle approach to explore whether the educational and methodological backgrounds influence a Nobelist’s recognition.

The remainder of the paper is organised as follows: Sect. 2 discusses the method, Sect. 3 reports the results, and Sect. 4 presents our concluding remarks.

Method and data

Nobel laureates are an obvious group of eminent scientists to study. By receiving the Nobel Prize, they have all achieved in their lifetime the highest accolade that one can receive as a scientist: “[I]t appears that the unplanned and tacit contest for prestige among awards is something like the planned and explicit contest of the decathlon in sport, with the Nobel emerging as champion through its high ranking in a variety of attributes making for prestige” (Zuckerman 1996, p. 20). From a quality perspective, Laureates can be seen as a homogenous group of scientists who have achieved high quality work in their process of discovery and knowledge generation. We are therefore able to hold quality relatively constant when exploring recognition throughout the career. Similarly, Zuckerman (1996, p. 62) finds that laureates are more productive early in their scientific career; they are rewarded quickly and are generally able to transform recognition into resources for future work. These common characteristics help to hold other factors as equal as possible.

Issues related to the previously mentioned selection effect (whereby better students self-select into better institutions) are less relevant when only studying Nobel laureates. In their youth, Nobelists have a good eye for finding a “master of their craft” (Zuckerman 1996, p. 109). Academia (and in particular Nobel laureates) supplies a “real-world laboratory” for testing how an individuals’ training background influences future success. In other labour market settings, such data is noisier and the job profiles, work goals and motivations are less comparable. We are able to quantify (via awards) individual career success with very little measurement error. Obviously, we have heterogeneity across disciplines as well as across the preferences of individuals. Laureates in the area of Physiology or Medicine have been able to generate more major awards (n = 1254) than Nobelists in Physics (938) and Chemistry (857). Zuckerman (1996) quotes a Nobel laureate to demonstrate the emotional challenges of awards and how people handle it differently: “[Baker, a pseudonym] was just over seventy when I went to his laboratory. A whole group went to his home and Mrs. Baker showed us all of his medals and there was something she said that made me realise that she was disappointed. It was undoubtedly a reflection of her husband’s feelings of disappointment that he had not been recognised by a Nobel award. Driving home with my wife, we got to talking about this and I said, ‘I am never going to worry or have a goal in mind of any prize, even a Nobel award. I refuse to die disappointed if I don’t get it.’ You put your happiness into the hands of some committee, which can be capricious. You’ve got to work for the fun of it. Men of equal accomplishment don’t get it and then they have to rationalise for the rest of their lives. But don’t get me wrong, I’m not sorry I got it” (pp. 209–210). However, incentives and constraints such as the “rules of the academic game” are clearly specified. Such conditions reduce omitted variable biases when conducting an empirical analysis.

In our analysis, we cover a 100-year period from 1901 to 2000, studying the achievements of all the Nobel laureates in physics, chemistry 12 and medicine or physiology. We therefore omit economics Nobel laureates. The data is derived from Kurian (2002) The Nobel Scientists: A Biographical Encyclopedia, a volume that provides very detailed information of other

12 For laureates who received the Nobel Prize twice, we use only the first Nobel Prize award [i.e. Marie Curie (Chemistry 1911), John Bardeen (Physics 1972) and Frederick Sanger (Chemistry 1980)].

13 major institutionalised awards obtained by these Nobel laureates. The advantage of this Encyclopedia is that it focuses only on very important prizes, leaving out awards such as honorary doctorates that could be perceived by some Nobelists or researchers as less important. For example, Zuckerman (1996, p. 224) reports how Nobel laureate Francis Crick devised a standardised checklist to deal with the flood of requests that he obtained. Going through the list we can see that he perceived honorary degrees as less important:

“Dr. Crick thanks you for your letter but regrets that he is unable to accept your kind invitation to:

send an autograph help you in your project

provide a photograph read your manuscript

cure your disease deliver a lecture

be interviewed attend a conference

talk on the radio act as chairman

appear on TV become an editor

speak after dinner write a book

give a testimonial accept an honorary degree.”

It is important to look at major recognitions, as the number of awards among Nobelists might be driven by the award culture of a country. Some countries have established more academic awards, and laureates can profit from such circumstances. Many of the major awards such as the Copley Medal, the Davy Medal, the Lasker Awards, the Enrico Fermi Award, the Franklin Medal, the Hughes Medal, the Max Planck Medal, the Gairdner Foundation International Award, or the Faraday Medal are available to scientists from different countries.

To measure the Nobelists’ educational background we look at the place and the country of their highest educational attainment. This

13 For further valuable resources on eminent people see Aubrey (1898/2007)s Lives of eminent men or Cattell’s (1921) American Men of Science biographical directory.

information is also available in Kurian (2002). In cases where researchers have obtained two doctoral degrees we take the institution where the laureate 14 obtained the most recent degree. Data to proxy a Nobelists’ methodological orientation are derived from Jones and Weinberg (2011), who used biographical sources to determine whether prize-winning work had an 15 important theoretical or empirical component. Such a proxy, although not perfect for our longitudinal focus, might be a good reflection of the overall methodological orientation throughout a scientist’s entire career. Our dataset indicates that 32.7 % of the laureates in Physics are theoretically oriented, while in Chemistry the share is 20.7 %. In the case of Physiology or Medicine, the number of laureates in our data set who are classified as theoretically oriented is small (13 out of 172). Thus, when analysing the disciplines independently we only focus on Chemistry and Physics.

Results

Descriptive analysis We first investigate the educational background of the Nobelists, noting the country of their highest education. Figure 3.1 shows the relative share of major awards through an academic career. As the number of Nobel Prize winners can vary between fields from year to year, we explore the number of awards per number of Nobel Prize winners within a field (number of awards in a particular year divided by number of laureates in that field based on laureates that are still alive). Moreover, we look at the number of awards relative to the number of Nobelists educated in a particular country. For example, the blue line (US) shows the number of awards divided by the number of Nobel laureates that obtained their highest education in the US. The figures report a five-year moving average window (smoothing). In Figure 3.2 we explore data on Nobelists who received the award before and after WWII separately. Before WWII, Germany was responsible for a large proportion of Nobelists (more than one-fourth), followed by Great Britain.

14 For example, Walter Kohn has PhDs from Harvard (1951) and Toronto (1954). We therefore classify him under Toronto and Canada. Alan G. MacDiarmid is another example with a Ph.D. from University of Wisconsin (1958) and from Cambridge (1961). He is therefore classified under Cambridge and Great Britain. 15 In 21 cases (out of 525 laureates) they observed a combination of theoretical and empirical work. These were classified under theoretical.

The US took the lead after the war (Zuckerman 1996), an effect that has been referred to as “Hitler’s Gift to American Science” (p. 69); whereby many eminent scholars left Germany after the Nazi government passed the “Law for the Restoration of the Professional Civil Service” in April 1933 (Waldinger 2012, p. 840). The list of important European researchers who moved to the US included, for example, Albert Einstein, Enrico Fermi, James Franck, Viktor Hess, Peter Debye, Otto Loewi, Otto Meyerhof, and Otto Stern. Nobel laureate Samuelson (2004) also points out: “Hitler gave us even before the war the cream of the continental crop” (p. 51). This provided a boost to the academic and educational system in the US. From our dataset on major awards, we can see that laureates educated in the US generated in total 1312 awards, followed by Great Britain (562) and Germany (364).

Figure 3.1 Awards and countries’ educational background

Figure 3.2 Nobel laureates who obtained the Nobel Prize before and after WWII

The figures demonstrate the tendency for Nobelists with a British education to generate more awards as they reach their peak recognition. On the other hand, researchers educated in the US receive more awards earlier in their career and are recognised later in their career. In addition, we observe that laureates with a German educational background receive their

40 recognition later in their academic career. These results are relatively stable over all three fields.

When performing a t test on the equality of the means using single 16 yearly values rather than moving average values we find that Nobelists educated in the US are more successful in generating major awards than all the other countries. Great Britain only dominates the US with respect to Physics, although the difference from the US is not statistically significant. Researchers educated in Great Britain are also more successful than those educated in Germany or the group of other countries. However, for Physiology or Medicine and Chemistry, the difference to Germany is not statistically significant. We then analyse the pre- and post-World War II period separately, based on the year when the Nobelists actually received their Nobel Prize (see Figure 3.2; Table 3.1). We observe that the differences between US and Great Britain remain statistically significant for both groups while the difference between Great Britain and Germany is no longer statistically significant for those Nobelists who received the Nobel Prize in Post-WWII period.

16 All the t-tests in this paper are conducted with single yearly values rather than moving averages.

Table 3.1 Educational differences

t-tests on the equality of the means by country of highest education All Great Britain Germany Other Countries United States 2.888*** 4.573*** 6.737*** Great Britain 1.782* 2.645** Germany -0.308 Chemistry United States 4.005*** 4.568*** 3.018*** Great Britain 0.771 -0.625 Germany -1.793* Physics United States -1.522 0.879 1.491 Great Britain 1.906* 2.878*** Germany 0.143 Physiology or Medicine United States 2.782*** 3.793*** 6.408*** Great Britain 1.241 2.441** Germany 0.278 Pre-WWII United States 1.894* 4.780*** 5.266*** Great Britain 2.892*** 3.950*** Germany 0.305 Post-WWII United States 2.537** 2.669*** 4.694*** Great Britain 0.318 0.602 Germany 0.082 Note: t-statistics for mean(Row – Column). The symbols *, **, *** represent statistical significance at the 10%, 5% and 1% levels.

Next, we look at the institution from which the laureates obtained their highest level of education. To represent the differences visually, we examine the most successful institutions based on the number of laureates generated, and put all other institutions into another category (Figure 3.3 and Figure 3.4). Cambridge has produced the largest number of Nobelists (9.81 %), followed by Harvard (5.76 %) and Columbia (4.05 %). Within Great Britain, Cambridge is responsible for having generated 60.5 % of all the Nobel laureates. Our data set also shows that Nobelists educated in Cambridge have attracted the largest number of major awards (356), followed by Harvard (211), University of California (172), and Columbia (148).

Figure 3.3 Institutions and award success

Figure 3.4 Institutions and award success before and after WWII

We also report the relative proportion of awards accrued to an institution by dividing the number of awards generated in a particular year by the number of laureates in a particular institution. The results indicate that these three institutions tend to dominate the number of awards obtained throughout the career of a scientist although the difference from the other universities is not statistically significant. Moreover, the difference between the three top places and the group of all the other universities is driven by the

pre-WWII period. For Nobelists who obtained the Nobel Prize after WWII the differences between these institutions are no longer statistically significant. When we look at all the fields together, the difference between Cambridge, Harvard and Columbia is not statistically significant (Table 3.2). However, we observe some heterogeneity between the fields. In the field of Chemistry, Cambridge and Harvard dominate Columbia. In Physics, Columbia and Cambridge dominate Harvard. In Physiology or Medicine, Columbia performs best but the difference between the three institutions is not statistically significant. In Chemistry, there is no statistically significant difference between the other universities and Cambridge or Harvard, while Columbia actually performs worse. For Physics, laureates educated in Cambridge and Columbia perform better than laureates with their highest degree from other universities. Harvard dominates the other institutions in Physiology or Medicine. Pre-WWII, laureates educated in Columbia are more successful than Harvard, Cambridge and the group that covers all other institutions.

Table 3.2 Educational institutional differences

t- tests on the equality of the means by Top University All Harvard Columbia Other Universities Cambridge 2.888*** 4.573*** 6.737*** Harvard 1.782* 2.645** Columbia -0.308 Chemistry Cambridge 4.005*** 4.568*** 3.018*** Harvard 0.771 -0.625 Columbia -1.793* Physics Cambridge -1.522 0.879 1.491 Harvard 1.906* 2.878*** Columbia 0.143 Physiology or Medicine Cambridge 2.782*** 3.793*** 6.408*** Harvard 1.241 2.441** Columbia 0.278 Pre-WWII Cambridge 1.894* 4.780*** 5.266*** Harvard 2.892*** 3.950*** Columbia 0.305 Post-WWII Cambridge 2.537** 2.669*** 4.694*** Harvard 0.318 0.602 Columbia 0.082 Note: t-statistics for mean(Row – Column). The symbols *, **, *** represent statistical significance at the 10%, 5% and 1% levels.

In a further step, we explore the difference between theoretically and empirically oriented researchers (see Figure 3.5 and Figure 3.6; Table 3.3). Overall, having a theoretical orientation leads to more awards throughout the career, particularly during the later stages. A comparison of the fields reveals that theorists are particularly successful in Chemistry. As mentioned in the methodological section we leave Physiology or Medicine out of the sub-field analysis. With respect to Physics, there is a difference between those laureates who received the Nobel Prize before WWII and those after the war. A theoretical rather than empirical orientation has a positive impact on awards among post-WWII Nobelists while the opposite is observed for pre-WWII laureates.

Figure 3.5 Empirical versus theoretical orientation

Figure 3.6 Pre- and post-award difference between sub-fields

Table 3.3 Theory versus Empirical Orientation

t-test on the equality of the means (by empirical versus theoretical) All -1.485 Chemistry -3.895*** Physics -1.129 Pre-WWII, Chemistry -1.121 Pre-WWII, Physics -1.195 Post-WWII, Chemistry -3.534*** Post-WWII, Physics 3.051*** Note: t-statistics for mean(Row – Column). The symbols *, **, *** represent statistical significance at the 10%, 5% and 1% levels.

Multivariate analysis We estimate the effect of our key variables by modelling the award count of

Nobel laureate i in year t (Ait) using a random-effects negative binomial regression model:

( + ) 1 Pr( = | , ) = , ( ) ( ) 𝜆𝜆𝑖𝑖𝑖𝑖 𝑎𝑎𝑖𝑖𝑖𝑖 𝑖𝑖𝑖𝑖 𝑖𝑖𝑖𝑖+ 1 1 + 1 +𝑖𝑖 𝑖𝑖𝑖𝑖 𝑖𝑖𝑖𝑖 𝑖𝑖𝑖𝑖 𝑖𝑖 Γ 𝜆𝜆 𝑎𝑎 𝛿𝛿 𝐴𝐴 𝑎𝑎 𝑋𝑋 𝛿𝛿 𝑖𝑖𝑖𝑖 𝑖𝑖𝑖𝑖 � 𝑖𝑖� � 𝑖𝑖� where Γ(∙) denotes the gammaΓ 𝜆𝜆integral,Γ 𝑎𝑎 = (𝛿𝛿 ), t is 𝛿𝛿a vector of

𝑖𝑖𝑡𝑡 𝑖𝑖𝑡𝑡 𝑖𝑖 individual-specific characteristics, and i𝜆𝜆 is the𝑒𝑒𝑥𝑥 𝑝𝑝dispersion𝑋𝑋 𝛽𝛽 𝑋𝑋 parameter that varies randomly across individuals with 𝛿𝛿1/(1 + ) ~ ( , ). Unlike the 𝑖𝑖 Poisson regression, this model is designed 𝛿𝛿to explicitly𝐵𝐵𝑒𝑒𝑡𝑡𝑎𝑎 𝑟𝑟 𝑠𝑠handle over- 17 dispersion, which has been tested for and is a feature of our data. As independent variables we include a dummy for being theoretically oriented (Theoretical Orientation), dummies for educational background based on the highest educational attainment (Great Britain, US, Germany, and Other Countries (reference group)), a quadratic time trend (Years since Highest Educational Degree, Years since Highest Educational Degree^2), a dummy for the Post Nobel Prize Period based on our former results (Chan et al. 2014b), dummies for the field where a scientist obtained the Nobel Prize (Chemistry, Physics, Physiology or Medicine (reference group)), and a dummy for those laureates who obtained the Nobel Prize after WWII (Post WWII).

In Table 3.4 we present the first two specifications including only our main variables. In specification (1) we restrict the sample to laureates who have already died. In the next two specifications we report the full set of control variables (see (3) and (4)). In the final two specifications we exclude the Nobel Prize in the dependent variable (specification (5) and (6)).

In line with results from the descriptive section, the regression results strongly suggest that theoretically oriented laureates are receiving more awards than empiricists: the coefficient on Theoretical Orientation is statistically significant. The estimated marginal effect of Theoretical Orientation on the number of awards indicates that theoretically oriented Nobelists receive on average between 0.017 and 0.035 more awards per year

17 For a discussion on the test see Cameron and Trivedi (2009, p. 561).

47 than empirically oriented Nobelists. When looking at those Nobelists who have died we observe that the period between finishing their highest education and their death is on average 53 years. Thus, taking specification (3) as an example, this would mean that theoretically oriented laureates generate on average 1.86 more major awards than those with an empirical orientation, an effect that cannot be ignored. It should be noted that on average these Nobel laureates generate 6.85 major awards in their career.

Table 3.4 The impact of educational and methodological background on awards using random effects negative binomial model

Dependent Variable: Without NP Without NP Number of Major Awards Died NB All Died NB All NB Died NB All Independent Variables (1) (2) (3) (4) (5) (6) Theory versus Empiricism Theoretical Orientation 0.244*** 0.135** 0.252** 0.156* 0.253** 0.183* (3.210) (1.986) (2.497) (1.663) (2.374) (1.870) 0.030 0.017 0.035 0.025 0.028 0.022 Educational Background Great Britain 0.326*** 0.189** 0.386*** 0.275** 0.393*** 0.254** (3.589) (2.253) (3.295) (2.432) (3.189) (2.175) 0.040 0.024 0.054 0.043 0.043 0.030 US 0.417*** 0.274*** 0.396*** 0.300*** 0.442*** 0.295*** (5.212) (4.042) (3.752) (3.199) (3.981) (3.049) 0.051 0.035 0.055 0.047 0.048 0.035 Germany -0.176* -0.226** -0.371*** -0.305** -0.388*** -0.377*** (-1.839) (-2.544) (-3.023) (-2.547) (-2.946) (-3.006) -0.022 -0.029 -0.052 -0.048 -0.042 -0.045 Other Countries Ref. Ref. Ref. Ref. Ref. Ref. Time Element 0.162*** 0.160*** 0.140*** 0.139*** Years Since Highest Educational Degree (21.867) (25.521) (18.175) (21.380) 0.023 0.025 0.015 0.017 -0.002*** -0.002*** -0.002*** -0.002*** Years Since Highest Educational Degree^2 (-17.304) (-19.080) (-15.652) (-17.350) 0.000 0.000 0.000 0.000

Table 6.4 continued

Dependent Variable: Without NP Without NP Died NB All Died NB All Number of Major Awards NB Died NB All Independent Variables (1) (2) (3) (4) (5) (6) Field Chemistry 0.181* 0.073 0.161 0.061 (1.765) (0.773) (1.489) (0.621) 0.025 0.012 0.017 0.007 Physics 0.088 0.050 0.045 -0.022 (0.878) (0.545) (0.423) (-0.236) 0.012 0.008 0.005 -0.003 Physiology or Medicine Ref. Ref. Ref. Ref. Period Post Nobel Prize Period -1.391*** -1.685*** -0.743*** -1.056*** (-17.254) (-23.682) (-9.210) (-14.985) -0.194 -0.265 -0.081 -0.126 Post World War II -0.209** -0.125 -0.033 0.012 (-2.365) (-1.434) (-0.359) (0.136) -0.029 -0.020 -0.004 0.001 N 17959 23404 17959 23404 17959 23404 Prob.>χ2 0.000 0.000 0.000 0.000 0.000 0.000 Notes: NP Nobel Prize. Coefficients in bold, z-statistics in parentheses and marginal effects in italics. The symbols *, **, *** represent statistical significance at the 10%, 5%, and 1% levels, respectively. Number of Nobelist=465 in total and 340 for those who have already died (specification (1), (3) and (5).

Being educated in Great Britain and the US also increased the number of awards, while Nobelists with a German educational background generate fewer awards than laureates educated in other countries (reference group).

With respect to Years Since Highest Educational Degree, we see a non-linear relation (increasing at a decreasing rate) as represented in the previous figures. Field differences do not, ceteris paribus, matter. However, as reported by Chan et al. (2014b, we observe that after researchers receive the Nobel Prize they attract significantly fewer awards. There is a trend that researchers who became laureates after WWII generate fewer awards. However, in most of the specifications the coefficient is not statistically significant.

In Table 3.5 we explore the educational background of the Nobelists, focusing on the institution rather than the country and using the same specification structures reported in Table 3.4 specifications (3) to (6). The results show that laureates educated in Cambridge perform best. With an average post educational period of 53 years, they are able to obtain 3.07 more major awards than the laureates who obtained their highest education from the universities in the reference group (specification 3B). Changing the reference group we also find that the differences between Cambridge, Harvard, and Columbia are not statistically significant.

Table 3.5 Institutional Educational Background and Award Success

(3B) (4B) (5B) (6B) University of Cambridge 0.410*** 0.339*** 0.353*** 0.437*** (3.185) (2.708) (2.728) (3.218) 0.058 0.055 0.043 0.048 Harvard University 0.243 0.084 0.146 0.336* (1.318) (0.534) (0.903) (1.730) 0.035 0.014 0.018 0.037 Columbia University 0.177 0.027 0.091 0.290 (0.873) (0.146) (0.481) (1.356) 0.025 0.004 0.011 0.032 N 17989 23434 23434 17989 Notes: Regression specifications based on the former table, exchanging the countries’ educational background with the institutional background. All the other universities are in the reference group. Coefficients in bold, z-statistics in parentheses and marginal effects in italics. The symbols *, **, *** represent statistical significance at the 10%, 5%, and 1% levels, respectively.

Conclusions

Awards have emerged as a symbol of prestige and scientific standing. Receiving a major academic award is a signal to the recipients that they have contributed to the advancement of knowledge and to the academic empire. In an environment where accomplishment is not easy to measure, awards represent success. Awards also enhance access to resources that can be used to cement and promote research quality (Merton 1968; Zuckerman 1996). Over the past few years, the number of scientists has increased significantly and so has the number of awards (Merton 1969; Stephan 2012). It is therefore valuable to explore how educational and methodological backgrounds shape researchers’ future recognition. Analysing only Nobel laureates allows us to hold individual research quality relatively constant when exploring recognition throughout a researchers’ career. However, we take advantage of the fact that their level of recognition differs throughout their careers. Future research could also explore in more detail the career differences between nominated scientists and actual award winners. For example, Rablen and Oswald (2008) find that Nobel Prize winners go on to have longer lives than scientists who were nominated. In those cases where nominations are not available one can develop a synthetic control group of nonrecipient scientists with similar pre-award research performance. Chan et al. (2014a) used such an approach to explore recipients of the John Bates Clark Medal and appointees to the Fellowship of the Econometric Society. They observe significant positive publication and citation differences after award receipt. Azoulay et al. (2014) explore the event of life scientists’ appointment as investigators at the Howard Hughes Medical Institute (HHMI) to determine the effect of status changes on researchers’ citation performance. However, they find only a moderate post-appointment citation boost for articles published before the appointment.

The skills and knowledge generated as students can have long-lasting career implications. We observe that laureates with a theoretical orientation are able to acquire more awards than empirically oriented Nobelists. On average, a Nobelist who has already died generates 6.85 major awards throughout his/her life. The average length of life after finishing the highest educational achievement is 53 years. Throughout this period, a theoretically oriented laureate is able to generate on average around 1.86 more awards than the empirically oriented laureate. Having received the highest level of

52 educational attainment from the US and Great Britain is beneficial for academic recognition. There is a tendency for Nobelists with a British education to generate more awards when experiencing their peak award success while researchers educated in the US receive more awards both early and late in their career. Future research could consider a close investigation into whether such results are driven by the availability of the amount of awards.

In addition, we observe that laureates educated in Germany obtain fewer awards and receive recognition later in their academic career. At the institutional level we find that laureates educated at the University of Cambridge and Harvard University generate more recognition than other laureates in Chemistry, while laureates educated in Columbia and Cambridge dominate in Physics. Harvard educated laureates are more recognised than others in Physiology or Medicine.

Cambridge, for example, with its argumentative tradition and its open door policy has stimulated scientific excellence (Erren 2008). Zuckerman (1996) refers to the Cavendish Laboratory at the University of Cambridge as a good example of an intensive interaction between elite masters and excellent apprentices. As our results indicate, Cambridge performs quite well in the area of Physics where the Cavendish Laboratory is active. Eminent researchers provide a “bright ambiance” (a laureate’s statement provided in Merton 1968, p. 159). Close interactions and exchanges with leading researchers in a creative environment enhance human or intellectual capital. Being educated in a top place provides a good foundation for acquiring more of what is required for future success and therefore in achieving recognition.

Chapter 4 Do Nobel Laureates Change Their Patterns Of Collaboration Following Prize Reception?

Chan Ho Fai, Ali Sina Önder & Benno Torgler

Scientometrics (2015), 105(3), 2215-2235.

Abstract

We investigate whether Nobel laureates’ collaborative activities undergo a negative change following prize reception by using publication records of 198 Nobel laureates and analysing their coauthorship patterns before and after the Nobel Prize. The results overall indicate less collaboration with new coauthors post award than pre award. Nobel laureates are more loyal to collaborations that started before the Prize: looking at coauthorship drop-out rates, we find that these differ significantly between coauthorships that started before the Prize and coauthorships after the Prize. We also find that the greater the intensity of pre-award cooperation and the longer the period of pre-award collaboration, the higher the probability of staying in the coauthor network after the award, implying a higher loyalty to the Nobel laureate.

Introduction

Although the Nobel Prize is regarded as the highest scientific accolade (Zuckerman 1992), interviews with Nobel laureates (Zuckermann 1996, pp. 218–236) indicate that its reception can lead to disruptions and unintended consequences in the scientific work process because of the abrupt upward mobility it brings, and furthermore “the laureates’ relations with their collaborators change most decisively” (Zuckerman 1996, p. 232). In this paper we investigate changes in Nobel laureates’ collaboration patterns with their coauthors following the receipt of the Nobel Prize.

Several laureates have reported that the prize erected barriers of deference between themselves and their colleagues, separating them emotionally and putting a distance between them that is “sometimes transformed into envy and the inclination to remove the hero from his pedestal” (Zuckermann 1996, p. 231). The resulting reduction in effective communication and exchange can disrupt the Nobelist’s collaboration network, and thus lead to less interaction with collaborators. This is one of the possible ways a Nobelist’s collaboration patterns may change following the prize.

A Nobelist’s collaboration patterns may change due to several other factors, such as younger coauthors’ willingness to establish an independent reputation (Zuckerman 1996; Merton 1968), or Nobelists’ reduced concern with recognition, or Nobelists publishing less out of fear that the newer work might be judged as mediocre (Zuckerman 1996, 229). Nevertheless, some Nobelists choose to maintain their collaboration network in order to keep publishing, perhaps even compensating for the reduced research time caused by increased external activities and post-prize demands by taking advantage of the greater number of students who approach them. When Nobelists are keen to publish, they have an incentive to profit from collaborative work, bringing additional, complementary knowledge, skills and capacities to a research project. As is often argued, the result of collaboration is more than the sum of the single parts: “[W]hen Watson and Crick set out to author an article together, a new author emerged, one not completely reducible to the two individual authors, James Watson and Francis Crick” (Wray 2006, p. 510). Not only has diversity of perspectives always been crucial to science (Shaman et al. 2013), but collaboration often emerges when the challenge at hand cannot be tackled by a single person (van Rijnsoever and Hessels 2011).

Laureates may also maintain networks because they are accustomed to, and reluctant to deviate from, certain habits of publication frequency. In such cases, network stability can be strengthened by a desire or willingness to mentor young scientists, while collaborative possibilities are heightened by the Nobelist’s ability to attract grants and greater access to scientific personnel and money. Collaboration can thus be either a strategic choice (Bozeman and Corley 2004) or one driven by curiosity or the shared excitement of conducting research and experiencing intellectual companionship (Heinze and Kuhlmann 2008; Beaver 2001; Katz and Martin

1997). Continued research effort may also be inspired by positive feedback on quality publications and a reputation for still being active after receiving the Nobel Prize. The net effect is thus hard to predict.

In addition to Zuckerman’s (1996) detailed analysis, several other studies focus on Nobel laureates, taking into account such factors as age or career path and productivity (Jones and Weinberg 2011; Kademani et al. 2005; van Dalen 1999; Stephan and Levin 1993), intuition (Marton et al. 1994), recognition across the career (Chan, Gleeson and Torgler 2014), speed of post-prize recognition (Chan and Torgler 2013), the consequences of educational background and methodological orientation (Chan and Torgler 2015), age premium (Baffes and Vamvakidis 2011), case study analysis of collaboration structure (Kademani et al. 2005), collaboration productivity (Chan et al. 2016), family background (Rothenberg 2005), professional ability (Shavinina 2004), predictability of the Nobel Prize (Gingras and Wallace 2010) and knowledge spillover (Ham and Weinberg 2011). In general, the exploration of Nobelists offers several advantages similar to those of a controlled (experimental) environment in that all prize winners have been affected by the same abrupt upward mobility shock and all are researchers with very high intellectual human capital and are thus relatively homogeneous in their collaboration “attractiveness”.

We explore whether coauthorship patterns of Nobel laureates experience a change upon prize reception by analysing the Nobel laureates’ collaboration patterns with their coauthors before and after the award. Specifically, we identify when and how many new coauthors join and leave the Nobelist’s collaboration network, measure the dropout rate of coauthors who collaborated with the laureate before the Nobel Prize, and assess whether these dropout rates are negatively correlated with collaboration intensity in the pre-award period. Our study thus contributes to the literature on scientific careers, which grew out of questions related to the skewed distribution of research productivity among scientists (Börner et al. 2010; Stokols et al. 2008; Dietz and Bozeman 2005).

A descriptive analysis of collaboration trends

We explore the award’s collaborative implications by carefully analysing all pre and post award publications of 198 Nobel laureates listed in SCOPUS, 18 whose records cover papers published between 1923 and 2014. Our sample comprises 1970–2000 Nobel laureates in physics (N = 71), chemistry (N = 56), and medicine or physiology (N = 71) and thus excludes two-time winners John Bardeen (1956 and 1972) and Frederick Sanger (1958 and 1980). A total of 34,287 co-authored publications are included in the analysis, of which 13,095 publications are co-authored by chemistry laureates; 6,959 are publications by physics laureates; and 14,233 by medicine or physiology laureates.

Our first focus of interest is whether a change occurs in arrival of new coauthors in the Nobel laureates’ collaboration networks after prize reception. Chan et al. (2016), for example, observe a nonlinear inverted U-shape relation between the number of new coauthors and laureate age, one that on average reaches a peak after age 60. Our results identify a positive trend in new collaborators in the period before the Nobel Prize, which changes abruptly after conferral, with yearly values fluctuating around the value observed at the time of the award (Figure 4.1). Because the exact time of the potential breakpoint is known, we use the Chow test (see Table 4.1) to identify a structural break—that is a strong enough (co-author) shift of the pre- and post- award slope. This procedure is equivalent to testing whether coefficients in two linear regressions comparing the period before and after the Nobel Prize are equal (fitted line in Figure 4.1), and it shows whether the rate of change of collaboration patterns differs before and after receiving the Nobel Prize. A statistically significant structural break at the time of prize receipt is confirmed for Nobel laureates in all age groups (age is defined here as the age of the laureate when she/he received the Prize), and in all three fields. Figure 4.1 illustrates the structural break, reflecting a negative change (from positive slope) in the rate of collaboration with new coauthors in the post award period. This structural break is even stronger when we restrict our sample to

18 We have compared the publication records from Web of Science with those from Scopus for a random sample of 50 Laureates. On average, Scopus has 6.14 more publication records than the Web of Science database for each of Laureate.

19 deceased Nobel laureates. In addition, we use a t test (see Table 4.1) for mean-comparison to assess the change in the level of pre- and post-prize collaboration measures, and the result of the t test indicates that Nobel laureates have, on average, more new coauthors after receiving the Prize.

Figure 4.1 Average of Nobelists’ new coauthors before and after the Nobel Prize

Note: green solid line is the linear fit for observations in the pre-award period (including the award year, t = 0). Blue dash line is the linear fit for the observations after the award year (from t = 1). Red line is set at 0.5 to distinguish before and after Nobel Prize.

19 Nobel laureates who passed away before 1st January, 2015.

Table 4.1 Chow test for structural breaks and mean comparison t-test: number of new coauthors

Number of new coauthors Chow test t-test DF F-stat. p-value DF Before NP After NP Diff. p-value

Full sample Nobel laureates in 1970-2000 6275 26.91 0.000*** 6277 5.01 7.95 -2.93 0.000*** Age for NP <= 46 747 3.91 0.020** 749 4.24 6.61 -2.37 0.000*** 46 < Age for NP <= 56 2246 4.01 0.018** 2248 4.95 10.23 -5.29 0.000*** 56 < Age for NP <= 66 1873 17.34 0.000*** 1875 4.99 6.04 -1.06 0.045** Chemistry 1965 13.08 0.000*** 1967 4.68 8.21 -3.53 0.000*** Physics 2025 13.42 0.000*** 2027 4.56 5.87 -1.32 0.048** Physiology or medicine 2277 8.11 0.000*** 2279 5.74 9.42 -3.69 0.000*** Deceased Nobel laureates Nobel laureates in 1970-2000 2852 16.08 0.000*** 2854 3.68 4.63 -0.95 0.012** Age for NP <= 46 132 0.61 0.542 134 1.83 3.74 -1.91 0.001*** 46 < Age for NP <= 56 638 3.91 0.021** 640 3.38 5.72 -2.35 0.000*** 56 < Age for NP <= 66 1091 5.53 0.004*** 1093 3.07 3.56 -0.48 0.31 Chemistry 894 15.5 0.000*** 896 3.58 3.96 -0.38 0.336 Physics 752 9.66 0.000*** 754 3.65 4.84 -1.19 0.326 Physiology or medicine 1198 2.96 0.052* 1200 3.77 5.03 -1.26 0.002*** Note: Deceased: data up to the end of 2014. *, **, *** represent statistical significance at the 10%, 5% and 1% levels, respectively.

Nobelists constitute a highly heterogeneous group with respect to age. We examine different age cohorts based on award year (Figure 4.2). In two of the three age categories (those who received the Prize before 47 and those who received between 57 and 66), we find that there is a negative trend in establishing new coauthorships after receiving the Nobel Prize. When we compare the results from different fields (Figure 4.3), a strong structural break (from a positive to a negative slope) is found for chemistry and physics. Overall, we observe a positive trend of collaborating with more coauthors before the Nobel Prize, and this trend turns negative after the prize, even though the average number of new coauthors is higher in the post-Prize period.

Figure 4.2 Different age cohorts based on the age at prize reception

Figure 4.3 Number of new coauthors by field

Next, we normalise the number of new coauthors by the size of the current coauthor network (see Figure 4.4). We observe that the trend for establishing new coauthorship is decreasing before the Nobel Prize, but it remains relatively stable afterwards. This pattern indicates a structural break (statistically significant at the 1 % level) similar to that shown in Table 4.2. In fact, when we test for the same set of subgroups, structural changes are observed in most cases, although less so for deceased laureates.

Table 4.2 Chow test for structural breaks and mean comparison t-test: entry rate

Entry rate Chow test t-test DF F-stat. p-value DF Before NP After NP Diff. p-value

Full sample Nobel laureates in 1970-2000 5881 6.55 0.001*** 5883 0.33 0.27 0.06 0.000*** Age for NP <= 46 667 17.16 0.000*** 669 0.43 0.3 0.14 0.000*** 46 < Age for NP <= 56 2147 4.26 0.014** 2149 0.37 0.29 0.08 0.000*** 56 < Age for NP <= 66 1778 0.93 0.397 1780 0.29 0.25 0.04 0.002*** Chemistry 1916 8.79 0.000*** 1918 0.33 0.27 0.06 0.000*** Physics 1777 1.6 0.202 1779 0.31 0.25 0.06 0.000*** Physiology or medicine 2180 5.7 0.003*** 2182 0.35 0.29 0.05 0.000*** Deceased Nobel laureates Nobel laureates in 1970-2000 2623 0.2 0.821 2625 0.3 0.24 0.06 0.000*** Age for NP <= 46 115 4.29 0.016** 117 0.44 0.27 0.17 0.003*** 46 < Age for NP <= 56 600 2.03 0.133 602 0.32 0.25 0.08 0.001*** 56 < Age for NP <= 66 1013 0.29 0.751 1015 0.29 0.23 0.05 0.002*** Chemistry 874 3.28 0.038** 876 0.3 0.23 0.08 0.000*** Physics 622 1.15 0.317 624 0.27 0.24 0.04 0.187 Physiology or medicine 1119 2.59 0.075* 1121 0.32 0.25 0.07 0.000*** Note: *, **, *** represent statistical significance at the 10%, 5% and 1% levels, respectively.

Figure 4.4 New coauthors divided by the number of existing coauthors

We move on to analysing the dropout of coauthors (termination of collaboration), and we first focus only on scientists who were collaborating with the laureate before conferral of the Nobel Prize. Based on ex post information, we are able to identify the year of the last cooperation, which is simply the last year of available publication data, and we take the year after the last collaboration as the termination year of collaboration. We can thus report separate results not only for deceased Nobel laureates but also for collaborators who began working with the laureate before the Prize. We observe a particularly strong increase in the number of pre-award coauthor dropouts in the 10-year period before the Prize, reaching the highest number of dropouts in the year of the prize (see

Figure 4.5 and Table 4.3). After that, the number of pre-award coauthor dropouts falls drastically until it is almost zero, which could imply that, even though we have no counterfactual to test the assumption, the prize itself may promote a high level of coauthor sustainability.

Table 4.3 Chow test for structural breaks and mean comparison t-test: number of pre-award coauthor dropouts

Number of pre-award coauthor dropouts Chow test t-test DF F-stat. p-value DF Before NP After NP Diff. p-value

Full sample Nobel laureates in 1970-2000 6272 228.7 0.000*** 6274 4.13 1.31 2.83 0.000*** Age for NP <= 46 747 40.16 0.000*** 749 2.97 0.76 2.21 0.000*** 46 < Age for NP <= 56 2243 127.68 0.000*** 2245 3.89 1.32 2.57 0.000*** 56 < Age for NP <= 66 1873 71.1 0.000*** 1875 4.22 1.1 3.13 0.000*** Chemistry 1965 72.16 0.000*** 1967 4.07 1.05 3.02 0.000*** Physics 2025 44.44 0.000*** 2027 3.58 1.28 2.3 0.000*** Physiology or medicine 2274 166.73 0.000*** 2276 4.71 1.55 3.16 0.000*** Deceased Nobel laureates Nobel laureates in 1970-2000 2852 64.79 0.000*** 2854 3.25 0.95 2.29 0.000*** Age for NP <= 46 132 8.37 0.000*** 134 1.26 0.44 0.82 0.024** 46 < Age for NP <= 56 638 49.26 0.000*** 640 2.85 0.75 2.11 0.000*** 56 < Age for NP <= 66 1091 31.45 0.000*** 1093 2.59 0.83 1.76 0.000*** Chemistry 894 25.69 0.000*** 896 3.23 0.81 2.42 0.000*** Physics 752 12.39 0.000*** 754 3.35 0.97 2.38 0.003*** Physiology or medicine 1198 55.78 0.000*** 1200 3.18 1.05 2.12 0.000*** Note: *, **, *** represent statistical significance at the 10%, 5% and 1% levels, respectively.

Figure 4.5 Number of pre-award coauthor dropouts

The stability is also driven, however, by the fact that fewer pre-award coauthors are still collaborating, some having already left the network. Thus, 20 we inspect the pre-award coauthor dropout rate. We observe a relatively stable dropout rate at around 20 %, on average (e.g., 1 out of 5 coauthors stops collaborating every year), before the Nobel Prize, and a decreasing dropout rate after the Prize, indicating higher level of collaboration sustainability (see Figure 4.6 and Table 4.4).

Next we explore potential sources of heterogeneity using different age cohorts (see Figure 4.7) as well as field (Figure 4.8). When our calculations are based on number of coauthor dropouts, we do observe a post- prize structural break (see also Table 4.3); however, when they are based on dropout rate (Table 4.4), we do not observe a structural break for the youngest age cohorts and in physics (and chemistry to a lesser extent).

20 Dropout rate is measured by the number of pre-award coauthors’ dropouts divided by the number of current pre-award coauthors.

Figure 4.6 Dropout rates for pre-award coauthors

Figure 4.7 Dropout rate by age cohort

Figure 4.8 Dropout rate by field

Table 4.4 Chow test for structural breaks and mean comparison t-test: dropout rates for pre-award coauthors

Number of pre-award coauthor dropouts Chow test t-test DF F-stat. p-value DF Before NP After NP Diff. p-value

Full sample Nobel laureates in 1970-2000 5705 13.07 0.000*** 5707 0.21 0.13 0.08 0.000*** Age for NP <= 46 602 2.46 0.086* 604 0.23 0.12 0.11 0.000*** 46 < Age for NP <= 56 2086 8.75 0.000*** 2088 0.22 0.12 0.1 0.000*** 56 < Age for NP <= 66 1731 10.87 0.000*** 1733 0.19 0.13 0.07 0.000*** Chemistry 1869 5.4 0.005*** 1871 0.22 0.12 0.1 0.000*** Physics 1713 0.73 0.481 1715 0.2 0.12 0.09 0.000*** Physiology or medicine 2115 10.78 0.000*** 2117 0.21 0.14 0.07 0.000*** Deceased Nobel laureates Nobel laureates in 1970-2000 2548 7.58 0.001*** 2550 0.21 0.12 0.09 0.000*** Age for NP <= 46 115 0.73 0.484 117 0.22 0.11 0.11 0.039** 46 < Age for NP <= 56 569 5.87 0.003*** 571 0.21 0.11 0.09 0.000*** 56 < Age for NP <= 66 970 6.5 0.002*** 972 0.2 0.13 0.07 0.000*** Chemistry 845 2.63 0.072* 847 0.21 0.12 0.09 0.000*** Physics 611 0.23 0.796 613 0.2 0.13 0.07 0.005*** Physiology or medicine 1084 6.81 0.001*** 1086 0.21 0.12 0.09 0.000*** Note: *, **, *** represent statistical significance at the 10%, 5% and 1% levels, respectively.

So far, we calculate dropout rates based on coauthorship relations that were established prior to the reception of the Prize, demonstrating that dropout rates of pre-award collaborations significantly decrease in the years following the receipt of the Nobel Prize. That is, collaborations initiated before the Prize are less likely to be dropped if these collaborations survive the receipt of the Prize. In order to obtain a more complete picture, we re- define the dropout rate; that is, we divide the number of ‘dropped out’ coauthors by the number of current coauthors regardless of the start of the collaboration. Using such re-definition, we obtain a structural break only for the overall sample (statistically significant at the 5 % level), but no significant structural break for separate age cohorts and fields (Table 4.5); moreover, in this case, the slope changes from negative in the pre-award period to positive in the post-award period (Figure 4.9). In addition, no structural break is observed for the subsample of deceased laureates, and the pre- and post-award levels of dropout rates are not significantly different except for physiology or medicine and the 56–66 age cohort albeit only by 3%. One possible interpretation of these findings regarding dropout rates would be that there is no significant difference between the collaborations initiated and ended before the Prize, and the collaborations started and ended after the Prize with respect to how these collaborations finish. Our analysis of pre-award dropout rates reveals that a core subset of collaborators survive the receipt of the Prize and face a significantly lower probability of being dropped; moreover, this is not because universal dropout patterns change for Laureates after the Prize, rather because this core group of collaborators are those deemed “vital” by the Laureate.

Table 4.5 Chow test for structural breaks and mean comparison t-test: dropout rates for all coauthors

Natural dropout rates for post-award coauthors Chow test t-test DF F-stat. p-value DF Before NP After NP Diff. p-value

Full sample Nobel laureates in 1970-2000 5985 2.68 0.069* 5987 0.21 0.22 -0.01 0.042** Age for NP <= 46 686 1.29 0.276 688 0.23 0.22 0.00 0.863 46 < Age for NP <= 56 2179 1.04 0.354 2181 0.22 0.23 0.00 0.637 56 < Age for NP <= 66 1799 0.35 0.701 1801 0.19 0.22 -0.03 0.005*** Chemistry 1933 2.21 0.11 1935 0.22 0.22 -0.01 0.529 Physics 1836 2.14 0.118 1838 0.20 0.20 0.00 0.877 Physiology or medicine 2208 0.07 0.933 2210 0.21 0.24 -0.03 0.002*** Deceased Nobel laureates Nobel laureates in 1970-2000 2675 0.61 0.545 2677 0.21 0.21 0.00 0.953 Age for NP <= 46 120 0.96 0.384 122 0.22 0.20 0.02 0.693 46 < Age for NP <= 56 609 1.19 0.306 611 0.21 0.21 0.00 0.975 56 < Age for NP <= 66 1027 0.12 0.886 1029 0.20 0.21 -0.02 0.187 Chemistry 880 0.1 0.904 882 0.21 0.20 0.02 0.241 Physics 648 1.66 0.191 650 0.20 0.20 0.00 0.847 Physiology or medicine 1139 0.09 0.911 1141 0.21 0.21 -0.01 0.533 Note: *, **, *** represent statistical significance at the 10%, 5% and 1% levels, respectively.

Figure 4.9 Dropout rates including post-award coauthors

Multivariate analysis

In this section, we conduct a multivariate analysis that estimates the time effects before and after the Nobel Prize. In those cases that explore the number of new coauthors or the number of dropouts, we use random effects negative binomial model that takes into account the individual heterogeneity of the laureates and the overdispersion in our data [see specifications (1) and (2) in Table 4.6]. The variance in the number of new coauthors (dropouts) is nearly 25 (15) times larger than the mean. When working with the dropout rate we use a simple random effects model [see specifications (3) to (6)]. As controls, we employ laureate age (age and square of age) to take into account a scientist’s career development, the gender of a scientist21, as well as research field. We also control for the nationality of Nobel laureates with a dummy variable (United States or other nationality), as it is possible that Non-US laureates are able to co-author more when they become better known after being awarded the Nobel Prize. For the time dummies, the reference period is the first 5 years after the Nobel Prize.

The results clearly show that the number of new coauthors increases after receipt of the Prize, but, all else being equal, the entry rate (new coauthors/current coauthors) is smaller in the post-award period. For example, the estimated marginal effect for 6–10 years after the prize on the number of new coauthors is 1.03 indicating that in this time period, the laureate has on average one more collaborator per year compared to first 5 years after receipt of the Prize [specification (1)]. The new entry rate, however, indicates that differences before and after the 5-year post-award period are not statistically significant [specification (3)]. The number of dropouts [specification (2)] is also smaller for the periods 6–20 years after prize reception compared to directly after the Nobel Prize. Only the period 5 years before the award shows a larger number of dropouts (statistically significant at the 1 % level), an average of 1.95 per year more in relation to the reference period. These results remain robust when considering the pre- award coauthor dropout rates [specification (4)] except that the outcome for the period 11–20 years before the prize is no longer statistically significant.

21 An abundant literature has shown significant gender difference in research collaboration structure; see Abramo et al. (2013) for a recent review.

Finally, the result for the dropout rate that also takes into account post-award coauthors are not significant for each period before and after the Prize.

For the field and age controls, we observe that physics shows more stability than the physiology/medicine control group, reporting fewer new coauthors and also fewer dropouts. The age at which the Nobel laureate receives the prize also matters in terms of arrival and dropout of coauthors: younger scholars are more susceptible to dropout than more senior researchers yet they tend to collaborate with more new researchers. In addition, by regressing the number of new co-authors on the interaction terms of the time dummies and nationality dummy, we find that US laureates have significantly more new co-authors in the periods before receiving the Nobel Prize; a significance that disappears in the post-award periods.22

22 Results are available upon request.

Table 4.6 Effects before and after the Nobel Prize

Negative Binomial (NB) Negative Binomial (NB) GLS Random Effects GLS Random Effects GLS Random Effects Random effects Random effects Dep. var. # new coauthors # coauthor dropouts Entry rate Dropout rate Dropout rate (# new coauthors (# pre-award coauthor dropouts (# coauthor dropouts / # current coauthors) / # current pre-award coauthors) / # current coauthors) Indep. var. (1) (2) (3) (4) (5) 16 to 20 years before Prize -.506*** -.521*** 4.7e-03 .028 .015 (.127) (.159) (.039) (.024) (.024) -5.85 -2.11 11 to 15 years before Prize -.462*** -.256** -.014 .028 8.0e-03 (.1) (.124) (.031) (.019) (.019) -5.34 -1.04 6 to 10 years before Prize -.253*** .011 -9.2e-03 .01 -.017 (.075) (.093) (.022) (.015) (.015) -2.93 .046 1 to 5 years before Prize -.108** .484*** -.01 .038*** 4.3e-03 (.053) (.063) (.015) (.012) (.012) -1.25 1.95 1 to 5 years after Prize (Ref.) (Ref.) (Ref.) (Ref.) (Ref.) 6 to 10 years after Prize .09* -.817*** -5.3e-03 -.058*** .01 (.054) (.08) (.016) (.013) (.012) 1.03 -3.3 11 to 15 years after Prize .061 -1.07*** -1.9e-03 -.038** 9.3e-03 (.077) (.115) (.023) (.016) (.016) .707 -4.32 16 to 20 years after Prize .151 -1.36*** .023 -.046** .019 (.106) (.167) (.032) (.022) (.021) 1.75 -5.51 Chemistry 6.7e-03 -.058 -.026 -.011 -4.1e-03 (.063) (.076) (.029) (1.0e-02) (9.9e-03) .099 -.294 Physics -.958*** -1.01*** -.036 -.015 -.018* (.063) (.074) (.028) (1.0e-02) (9.9e-03) -9.03 -3.35 Physiology or Medicine (Ref.) (Ref.) (Ref.) (Ref.) (Ref.)

Table 4.6 continued

Negative Binomial (NB) Negative Binomial (NB) GLS Random Effects GLS Random Effects GLS Random Effects Random effects Random effects Dep. var. # new coauthors # coauthor dropouts Entry rate Dropout rate Dropout rate (# new coauthors (# pre-award coauthor dropouts (# coauthor dropouts / # current coauthors) / # current pre-award coauthors) / # current coauthors) Indep. var. (1) (2) (3) (4) (5) Female -.678*** -.532** -.042 -.039 -.012 (.156) (.218) (.081) (.027) (.027) -7.83 -2.15 US nationality .104** -.012 -.027 -1.2e-03 1.4e-03 (.051) (.061) (.024) (8.2e-03) (8.1e-03) 1.2 -.05 Age for NP <= 46 .491*** -.357** .016 -6.5e-03 .02 (.148) (.181) (.053) (.025) (.024) 6.31 -1.25 46 < Age for NP <= 56 .158* -.19* .021 5.7e-03 .019 (.086) (.101) (.034) (.014) (.014) 1.71 -.722 56 < Age for NP <= 66 (Ref.) (Ref.) (Ref.) (Ref.) (Ref.) Age for NP > 66 .088 .374*** .07* .029* 8.5e-03 (.093) (.109) (.038) (.016) (.016) .92 1.9 Age .131*** .096*** -.018*** -3.1e-03 -9.9e-04 (.01) (.013) (2.7e-03) (2.1e-03) (2.0e-03) 1.51 .387 Age^2 -1.2e-03*** -1.1e-03*** 1.2e-04*** 1.8e-05 1.1e-05 (7.4e-05) (1.0e-04) (1.7e-05) (1.6e-05) (1.5e-05) -.014 -4.5e-03 Observations 6279 6276 5885 5709 5989 Number of Nobel laureate 190 189 190 189 190 LR χ2 821.0 1506.2 268.9 209.9 24.8 Prob > χ2 0.000 0.000 0.000 0.000 0.073 Notes: standard errors in parentheses. The symbols *, **, *** represent statistical significance at the 10%, 5% and 1% levels, respectively.

Next, we look for evidence of loyalty to (or of) the Nobelist by assessing whether more pre-award interactions are associated with a lower dropout probability of the pre-award coauthors (Table 4.7). We apply a probit model [specifications (7)–(8)] for the binary dependent variable loyalty (1 = have at least one publication together after the Prize reception). With respect to loyalty, we find that a longer pre-award collaboration history between laureates and their coauthors as well as a greater number of pre-award publications increases the probability that coauthors will not drop out of the network before the Nobel Prize. For example, 10 more pre-award publications above the average would raise the probability of staying in the network by 8.4 % points. Interestingly, our results also show that Nobelists who received the prize at quite a young age (under age 47) are more likely to maintain their collaborators compared to the reference group who received it between the ages of 47 and 56.

Finally, we explore whether pre-award and post-award collaboration intensity are positively correlated by using OLS [specifications (9)–(10)] and negative binomial regression models [specifications (11)–(12)]. We control for the length of the collaboration before the Nobel Prize, which allows us to hold it constant when exploring pre-award collaboration intensity. In addition to research field, we also measure the laureate’s age when the collaboration first started (in relation to each pair). Our regression results reveals that, all else begin equal, the extent of pre-award collaboration and collaboration length are positively correlated with the number of post-award collaborations. Our findings on preaward and post-award collaboration correlations as well as loyalty point to a very important characteristic of Nobelists: they seemed to know very well how to appreciate, nurture, and sustain collaboration within a productive and successful research team.

Table 4.7 Pre-award collaboration intensity and loyalty

Probit Probit OLS OLS NB Reg NB Reg Dep. var. Loyalty Loyalty Post NP Post NP Post NP Post NP collaboration collaboration collaboration collaboration Indep. var. (7) (8) (9) (10) (11) (12) Number of pre-NP collaboration .049*** .049*** .269*** .268*** .216*** .213*** (.018) (.018) (.087) (.087) (.022) (.022)

8.4e-03 8.5e-03 1.6e+20 1.0e+20 Collaboration start year (from NP year) .077*** .088*** .055*** .096*** .163*** .178*** (5.7e-03) (6.6e-03) (.011) (.024) (.016) (.018)

.013 .015 1.2e+20 8.6e+19 Chemistry .074 .086 .227 .158 .358* .324* (.082) (.08) (.165) (.148) (.194) (.19)

.013 .016 3.3e+20 1.9e+20 Physics -.105 -.103 .439 .38 .367 .338 (.097) (.096) (.448) (.422) (.238) (.245)

-.017 -.017 3.4e+20 2.0e+20 Physiology or Medicine (Ref.) (Ref.) (Ref.) (Ref.) (Ref.) (Ref.) Female -8.0e-03 5.5e-03 .595* .616* .714 .704 (.205) (.21) (.343) (.347) (.545) (.543)

-1.4e-03 9.5e-04 5.5e+20 3.4e+20 US nationality -.064 -.049 -.26 -.19 .173 .189 (.08) (.078) (.238) (.291) (.181) (.179)

-.011 -8.5e-03 1.3e+20 9.2e+19

Age for NP <= 46 .302** 1.37 .135

(.12) (1.08) (.347)

.058 1.4e+20

Table 4.7 continued

Probit Probit OLS OLS NB Reg NB Reg Dep. var. Loyalty Loyalty Post NP Post NP Post NP Post NP Indep. var. collaboration collaboration collaboration collaboration (7) (8) (9) (10) (11) (12)

56 < Age for NP <= 66 (Ref.) (Ref.) (Ref.)

Age for NP > 66 -.041 -.2 -.273

(.118) (.172) (.284)

-6.7e-03 -2.4e+20

Nobelist age at the first collaboration -.067*** -.052 .055

(.02) (.061) (.055)

-.012 2.7e+19

Nobelist age at the first collaboration2 5.6e-04*** 1.1e-04 -6.8e-04

(1.9e-04) (6.4e-04) (5.1e-04) 9.6e-05 -3.3e+17 Number of paired collaborations 21362 21362 21362 21362 21362 21362

Pseudo R2/R2 0.168 0.169 0.131 0.129 Prob. > χ2 /F/ χ2 0.000 0.000 0.000 0.000 0.000 0.000 Notes: Standard errors in parentheses. Marginal effects in italics. *, **, *** represent statistical significance at the 10%, 5% and 1% levels, respectively.

Conclusions

As Zuckerman’s (1996) interviews with Nobel laureates suggest, upward mobility does not always result in positive outcomes. Our results do in fact demonstrate a decreasing trend of new coauthors joining a Nobelist’s post award collaboration network. This finding is robust across most divisions of Nobelists based on age and field, with the only exception of laureates in physics, where there is no observable structural break. Our multivariate analysis suggests that the number of new coauthors increases after the Nobel Prize. With respect to the distance argument, we find no evidence that coauthors who were actively collaborating with the Nobel laureate before the award leave after the prize. On the contrary, not only does the dropout probability of pre-award coauthors decrease during the post award period, the number of dropouts increases quite substantially before the award. Considering the fact that the average team size producing a hard science publication has continuously and drastically increased over last several decades (Wuchty et al. 2007) any hint towards a decreasing trend of new coauthors or increase in dropouts in Nobel Laureates’ teams at any point in time turns out to be more striking.

The multivariate analysis further demonstrates that the dropout rates decrease 6–20 years after the Nobel Prize (relative to the 1–5 year post-award reference period). Once we include post-award coauthors, however, the dropout rates turn out not statistically different from that during 1–5 years after the Prize. The finding that the intensity of preaward publications and the length of pre-award collaboration history with the Nobelist reduces the probability of a coauthor leaving the laureate’s network implies that loyalty does matter.

This study is inherently descriptive because we offer no counterfactual such as a control group of scientists with similar coauthor structure and pre-award development to compare with the Nobelists over time. Such a control group, however, although it would allow the inference of causal relationship between the variables and the award, would be extremely difficult to find. One approach might be to look at scientists who were nominated as laureates but did not receive the prize, a list of whom is

23 provided by the Nomination Archive (albeit currently only up to 1963, which is useless for our 1970–2000 dataset). Moreover, even when focusing on nominees, we cannot assume that their coauthor network patterns are similar in the pre-award period, and substantial differences make comparison even more difficult. This current study contributes to this approach by suggesting an important first step in identifying possible Nobel Prize effects; namely, the use of a Chow test to identify structural breaks. Future studies could thus take these insights as a starting point for generating more precise size effects of being awarded the Nobel Prize.

23 http://www.nobelprize.org/nomination/archive/.

Chapter 5 The First Cut is the Deepest: Repeated Interactions of Co-Authorship and Academic Productivity in Nobel Laureate Teams

Chan Ho Fai, Ali Sina Önder & Benno Torgler

Scientometrics (2016), 106(2), 509-524.

Abstract

Despite much in-depth investigation of factors influencing the coauthorship evolution in various scientific fields, our knowledge about how efficiency or creativity is linked to the longevity of collaborative relationships remains very limited. We explore what Nobel laureates’ coauthorship patterns reveal about the nature of scientific collaborations looking at the intensity and success of scientific collaborations across fields and across laureates’ collaborative lifecycles in physics, chemistry, and physiology/medicine. We find that more collaboration with the same researcher is actually no better for advancing creativity: publications produced early in a sequence of repeated collaborations with a given coauthor tend to be published better and cited more than papers that come later in the collaboration with the same coauthor. Our results indicate that scientific collaboration involves conceptual complementarities that may erode over a sequence of repeated interactions.

Introduction

Dramatic changes in science over the past decades have increased task complexity, reshaping how scientists cooperate and turning science into a team effort (Katz and Martin 1997; Adams et al. 2005). In particular, the one- author-per-paper trend that dominated science from the 1600s until around the 1920s decreased in the 1950s, was barely visible by the 1980s (Greene

2007), and has become a rarity in scientific journals today. For example, of the 700 reports published in Nature in the first 10 months of 2008, only six were single author papers (Whitfield 2008). Our understanding of such collaboration is informed by visualisation of collaborative patterns (Newman 2004) and an evolving understanding of the principles of team formation (Guimera et al. 2005; Milojević 2014), which provides useful insights into optimal team size. The emerging use by scientists of collaborative indexes to more effectively measure researchers’ scientific impact (Stallings et al. 2013) also suggests that in the past few decades, single authors have performed worse than teams (Wuchty et al. 2007). Nevertheless, knowledge of how teams perform over time remains limited.

To help fill this void, we explore the productivity patterns of repeated scientific collaborations by Nobel laureates and their collaborators, thus ensuring a homogenous focus group of productive scientific “stars” with intellectual human capital of extraordinary scientific value. In particular, laureates are homogenous in their capacity to produce successful, innovative ideas and attract fairly able co-authors (Zuckerman 1996), which allows us to focus on team efficiency while holding team talent constant.

Data and descriptive analysis

Our dataset consists of 34,448 publications registered in Scopus (up to 2008) of 192 Nobel laureates who received the Nobel Prize in chemistry (56), physics (69), or physiology/medicine (67) between 1970 and 2000. The dataset includes 43,451 Nobel laureate coauthor pairs, for whose publications citation records are traceable up to 2014. The patterns of laureates’ accumulation of coauthors are similar in different fields. Although most Nobel laureates cooperate with fewer than 160 different coauthors over their academic lifecycle, a few cooperate with over 1000 different coauthors. The long tails of the histograms (Appendix Figure 5.4) somewhat reflect the fact that “hyper-authorships” tend to be the product of highly complex subfields such as biomedicine or high-energy physics (Cronin 2001).

Our first analysis explores the arrival of new coauthors and the intensity of coauthorship over Nobel laureates’ academic lifecycle. Figure 5.1 shows the number of new coauthors that appear in laureates’ publications at

82 a given age. The patterns for the arrival of new coauthors are comparable in chemistry, physics, and physiology/medicine before laureates reach age 60. Age 60 marks the peak for arrival of new coauthors in chemistry and physics, although there seems to be no clear peak in physiology/medicine.

Figure 5.1 Arrival of new coauthors by field

Note: Smoothed values are computed using restricted cubic spline.

The intensity of coauthorship captures the number of total collaborations between a laureate and a given coauthor (Figure 5.2). “Laureates’ Age” corresponds to the laureates’ age of first collaboration, with the vertical axis depicting the average number of collaborations between Nobel laureates and arriving coauthors (i.e., when collaboration begins) at a given age for the laureate. In chemistry and medicine, early collaborations tend to be more intense (albeit with a large variance). In physics, however, laureates’ intensities of coauthorship tend to show a positive trend at younger ages but no clear peak is observed.

Figure 5.2 Intensity of cooperation by field

Note: Smoothed values are computed using restricted cubic spline.

In addition, we refine the measure of collaborative intensity by taking into account the number of coauthors in each publication. For example, the level of collaboration intensity between coauthors on a publication with five coauthors may differ from the intensity experienced on a paper with two coauthors. We therefore utilise the A-index developed by Stallings et al. (2013) to account for each coauthor’s share in each publication. The A-index provides an estimation of the individual contribution (the relative share of credit among coauthors). Computation of the A-index requires grouping of coauthors according to their relative contributions to the publication. The groups are then ranked by the level of contributions. For the authors in the ith rank group, the A-index is defined as:

1 1 = 𝑚𝑚 ,

𝐴𝐴𝑖𝑖 � 𝑗𝑗 𝑚𝑚 𝑗𝑗=𝑖𝑖 ∑𝑘𝑘=1 𝐶𝐶𝑘𝑘 84

where m equals the total number of rank groups and ci is the number of coauthors in the ith rank group with the same level of contribution. The A- index is thus bounded by 1. To assign the rankings to each author based on the respective level of contribution, we follow Stallings et al. (2013) and Biswal (2013) in assuming that the listing order of the authors implies the relative contribution; that is, we assume the last author to be the corresponding author who has the same level of contribution as the first author (both ranked first), while the ranks for the other coauthors are in increasing order based on their listing (decreasing level of contribution). Table 5.3 shows the A-index calculated under this assumption for up to ten coauthors, although the A-index captures only the individual contribution. Thus, to measure the contribution of each Nobel laureate-coauthor pair, we propose the following method to calculate the collaboration contribution for a co-author pair using the A-index of author i and j:

= + , + 𝐴𝐴𝑖𝑖 ∗ 𝐴𝐴𝑗𝑗 𝐶𝐶𝑖𝑖𝑖𝑖 �𝐴𝐴𝑖𝑖 𝐴𝐴𝑗𝑗� ∗ 2 2 𝐴𝐴𝑖𝑖 𝐴𝐴𝑗𝑗 � � where Cij measures the co-contribution of author i and j with adjustment for the equity of the level of contribution between author i and j. The adjustment implies a larger discounting factor for coauthor-pairs with higher inequity with respect to the level of contributions between authors i and j. Thus, the

maximum value of Cij is equal to the sum of Ai and Aj. We make this adjustment because we assume that the intensity of collaboration between the coauthor-pair who contributed equally is higher than pairs with unequal

contribution given the same value of Ai plus Aj.

Figure 5.5 and Figure 5.6 show the weighted number of total collaborations between a laureate and a given coauthor assuming unequal and equal author contribution, respectively. The results resemble those in Figure 5.2 where early collaborations (before age of 40) are more intense.

We choose arrival of new coauthors and intensity of collaboration to capture the dynamics of Nobel laureates’ collaborations over their academic lifecycle because these reflect the social and academic norms in the respective fields. We assess the quality of such collaborations based on the number of citations received. For every laureate-coauthor pair that has published collaboratively in at least 4 distinct years, we calculate the average number of citations received by publications during first 2 years and last 2 years of collaboration. Figure 5.3 then plots the relationship between the two publication sets, with the average number of citations received by publications in first 2 years on the horizontal axis plotted against the average number of citations received by publications in last 2 years on the vertical axis (panel a). Panel b contains data restricted to laureate-coauthor pair that has published collaboratively in at least 7 distinct years. Data are plotted in the logarithmic scale. The red line represents the fitted values of a power law model between early and late citations (y = axb) and the green diagonal line indicates that late citations are equal to early citations (positively linear). The numbers of observations below and above the diagonal line are shown in the figure; the former (below the green) represents the number of coauthor-pairs where citations received by early publications are higher than citations for late publications, and vice versa for the latter. Results reveal that collaboration success is minimally dependent on pure luck: laureate and coauthor pairs that receive a high number of citations for their later publications are also those who receive a high number for their early publications (positive slope of the red line). Conversely, most collaborations that yield no highly cited publications early on tend to yield even fewer successful publications down the road.

Figure 5.3 Citations received by early and late collaborations of laureate- coauthor pairs

Note: The fitted values were obtained by linear least-square model, with the equation log10(y) = a + blog10(x). Data are plotted in the logarithmic scale.

The decay in citation success appears to be strongest in chemistry. The laureate coauthor- pair ratio for early citations to late citations is equal to 1.245 (799/642, see panel a), which indicates that early collaborations are more successful. The ratio in physics and physiology/medicine are similar (1.184 and 1.229 respectively). It is clear from panel b (representing more

long-term collaborations) that a greater number of observations lie below the diagonal line for chemistry (1.246) whereas more observations lie above the diagonal line in physics (ratio = 0.83) and physiology/medicine (ratio = 0.86), indicating indicate that late publications are more successful. The differences in citation success in earlier versus later publications over the lifecycle of a given collaboration is greater in chemistry, perhaps because most chemistry research is done in a way to generate very specific data that are best published within a few high impact publications. Research in physics and physiology/medicine, on the other hand, generate rather more multidimensional data that sustain a large number of good ideas leading to several high impact publications, especially in highly complex research areas where experiments require a very costly setup.

The results for citations adjusted by collaboration contribution

(citation counts multiplied by Cij) are depicted in Figure 5.7 and Figure 5.8 (for unequal and equal contributions, respectively). While the positive correlation between citations received by early and late publications remains robust when accounting for collaboration contributions, the ratio is mostly above 1 (with the exception of physics), which indicates a greater citation success for early publications.

Our results are robust to our definition of early and late, and they hold when we define early and late interactions to cover all interactions that fall into the first half and the second half of the collaboration period, respectively. These results are reported in Figure 5.9. For all disciplines, the ratio is above 1, indicating that the first period of collaboration is more successful than the second period. It is only for long-term collaborations (panel b) in physics that we observe the later period as more successful.

In order to investigate whether introduction of laureates who are still actively collaborating creates any bias in our analysis we differentiate between laureates who died before 2009 and those who are either still living or who died after 2009. The results are presented in Table 5.1, analysing laureate-coauthor pairs that have published in at least 4 distinct years. We provide an overview of the ratio results, which (in line with our initial analysis) focus on raw citation counts, citations weighted for equal and unequal co-author contribution, and an alternative definition of early and late collaborations as in the previous paragraph. Overall, we can see that the ratio

88 is mostly above one, indicating that early collaborations are more successful, which confirms the robustness of our initial results. The analysis of the deceased laureates further confirms the tendency in physics that later collaborations are more productive.

Table 5.1 Ratio of early to late citation success

Living laureates or laureates deceased after Laureates deceased before 2009 2009 n1 n2 Ratio n1 n2 Ratio Raw Citations counts (in line with Figure 5.3) Chemistry 453 604 1.333 189 195 1.032 Physics 1063 1254 1.180 158 192 1.215 Phy./Med 700 899 1.284 135 127 0.941 Citations weighted for unequal coauthor contribution (in line with Figure 5.7) Chemistry 386 671 1.738 172 212 1.233 Physics 1064 1253 1.178 209 141 0.675 Phy./Med 581 1018 1.752 123 139 1.130 Citations weighted for equal coauthor contribution (in line with Figure 5.8) Chemistry 389 668 1.717 178 206 1.157 Physics 997 1320 1.324 145 205 1.414 Phy./Med 617 982 1.592 125 137 1.096 First and second half of whole publication set (in line with Figure 5.9) Chemistry 394 663 1.683 181 203 1.122 Physics 1101 1216 1.104 215 135 0.628 Phy./Med 613 986 1.608 128 134 1.047 Notes: n1 represents the number of observations with more citations in the later publications than in the earlier publications and n2 represents the number of observations with more citations in the earlier publications than in the later ones. Ratio = n2 divided by n1.

Two-stage estimation and discussion of results

In the first stage (see Table 5.4), to isolate the correlation between citations received for an article and the intensity of cooperation between that article’s coauthors, we define journal quality as the journal’s 2012 impact factor from the ISI Web of Knowledge 2012 Journal Citation Reports and regress this variable on paper characteristics in the first stage estimation to obtain prediction errors ( , ). In the second stage estimation, we regress citation count on the same𝜇𝜇 explanatorŷ𝑖𝑖𝑖𝑖 ℎ variables as in the first stage but also on the predicted errors derived therein. The journal impact factor in the second step is thus the error obtained in the first, corresponding to the portion of journal

impact factor not explained by the paper and collaboration characteristics. In this way, we separate the effects of journal quality on citation success from other explanatory variables.

The bases for these estimations are the following two specifications:

Step 1: (Jouranl_Impact) , = (total collaboration, collaboration year,

𝑖𝑖𝑖𝑖 ℎ 𝑓𝑓#authors, laureate characteristics) + ,

𝜇𝜇𝑖𝑖𝑖𝑖 ℎ

Step 2: (Citations) , = (total collaboration, collaboration year,

𝑖𝑖𝑖𝑖 ℎ #authors,𝑓𝑓 laureate characteristics, , ) + ,

𝜇𝜇̂𝑖𝑖𝑖𝑖 ℎ 𝜀𝜀𝑖𝑖𝑖𝑖 ℎ We regress the journal impact factor for paper h of the laureate- coauthor pair ij on the total number of laureate (i) and coauthor (j) collaborations in our dataset (total collaboration), the year of appearance of that particular paper h in the life cycle of ij collaboration (in the first year, second year, or nth year of the collaboration), the total number of authors in publication h, and the Nobel laureate’s characteristics (field, age during publication, and individual fixed effects). To avoid collinearity between the total number of collaborations and the appearance number of a particular collaboration, we use indicator variables for various levels of total collaboration: 6–20, 21–40, 41–70, 71–110, and more than 110 (with between 1 and 5 as the reference group). Table 5.5 presents the descriptive statistics of the dependent and independent variables.

We focus on the marginal effects of repeated collaborations between laureate-coauthor pairs on citation success of their publications. In doing so, we must recognise that citations may be affected by the quality of the journal in which the article is published (e.g., due to increased visibility), or same variables affecting an article’s publication success may possibly be affecting also its citation success. Thus when citations are regressed on article’s characteristics that include publishing journal quality (measured by impact factor), such quality will be highly correlated with other explanatory variables. This correlation could produce misleading outcomes because journal quality and citation of the article, rather than being independent, are determined by the same exogenous factors, including collaboration intensity. The citation success results show that the first four collaboration bins are all

90 highly significant but negative (relative to the reference group of 5 or fewer collaborations), with only the fifth bin, the most extreme number of collaborations, being positive and insignificant (Table 5.2). Hence, all else being equal, and except for the extreme case of over 110 collaborations, the first cooperation sets tend to be more successful, leading to more citations per paper (between 16 and 48). Among laureates who won the prize while under 50, collaborations repeating more than 20 times have a positive and significant coefficient. For the laureates who won the prize after 50, the most successful papers are the early publications with the most intensive collaboration (over 110 repeated interactions). Most laureate-coauthor pairs collaborate over several years. The year (e.g., first, second, third …) of the laureate-coauthor collaboration in which a particular publication occurs is captured by the variable Collaboration Year in Table 5.2. Square of the collaboration year is included to capture the non-linear productivity pattern over the life cycle of collaborations. Long lasting collaborations are those that produce as good (or even better) cited publications during later years of collaboration as in the early years of it, and this is revealed by the non-linear marginal effect of the collaboration year. Non-linearity of citation success over the life cycle of a given collaboration captures an interesting relationship: although creativity and impact decays over the life cycle of many collaborations (most repeat over less than 4 distinct years), there are also some very long lasting collaborations that do not experience such a strong decay in productivity, hence the analysis should not be restricted to a strictly linear relationship between collaboration years and citation count.

Table 5.2 Regression results for the 2SLS

Overall Age of the Laureate Chemistry Physics Phy./Med <50 >=50 Collaborations 6-20 -15.65*** -11.95* -15.03*** 21.42*** -13.70*** 29.87*** (2.215) (6.949) (2.189) (4.424) (2.329) (3.476) Collaborations 21-40 -22.95*** 35.80*** -20.45*** 35.19*** -17.76*** 82.50*** (2.578) (8.654) (2.611) (7.995) (2.652) (5.024) Collaborations 41-70 -45.76*** 34.84*** -39.69*** 42.37*** -20.99*** 69.80*** (3.133) (7.823) (3.190) (11.007) (2.771) (5.736) Collaborations 71-110 -48.14*** 43.74*** -42.66*** 40.20*** -23.18*** 92.54*** (3.380) (9.877) (3.415) (11.945) (2.835) (7.378) Collaborations >110 4.87 62.15*** 4.15 26.97*** -1.68 76.61*** (3.173) (11.716) (3.132) (7.636) (2.848) (6.531) Collaboration Year -3.30*** -10.78*** -2.42*** -8.43*** -2.92*** -11.18*** (0.449) (1.577) (0.445) (1.211) (0.533) (0.974)

Collaboration Year^2 0.18*** 0.34*** 0.14*** 0.28*** 0.18*** 0.38*** (0.018) (0.058) (0.018) (0.041) (0.037) (0.037) Journal Quality 8.60*** 9.04*** 8.72*** 11.70*** 5.16*** 3.62*** (0.222) (0.478) (0.258) (0.682) (0.417) (0.175) Number of Authors 2.08*** 6.25*** 1.52*** -1.64 0.38*** 22.33*** (0.070) (0.325) (0.059) (1.184) (0.010) (0.511) Laureate Fixed Effects Yes Yes Yes Yes Yes Yes Time Fixed Effects Yes Yes Yes Yes Yes Yes Observations 155,179 27,961 127,218 32,473 77,016 45,690 Adjusted R-squared 0.347 0.446 0.345 0.161 0.212 0.674 Notes: The table reports second stage coefficients where the dependent variable is an article’s citation count. We run our analysis first for all Nobel laureates and then separately for laureates who won the prize while under or over the age of 50. We then conduct regressions for each field separately. Robust standard errors in parentheses. *p<0.1, **p<0.05, and ***p<0.01.

Comparing our results for different fields, we find that although the total number of citations received by a paper in chemistry and medicine is strongly positively correlated with the total number of collaborations between the laureate and that particular coauthor, earlier papers in the collaboration sequence are expected to receive higher citations. In physics, on the other hand, total number of citations is strongly negatively correlated with total collaborations, except for collaborations that repeat more than 110 times, thus most citations are expected for papers from collaborations that repeat either less than 5 times or more than 110 times.

Our results suggest a “collaborative idea scarcity”, meaning that ideas that come early in the lifecycle of a collaboration between coauthors are on average the most innovative ones based on citation count. This further suggests that a collaboration may run out of creative ideas over time. What, then, are the most likely reasons for such a result? One explanation may be that the creativity of the original combination that generates new insights and breakthrough may emerge early rather than later during researchers’ collaboration. Likewise, efficient problem solving may emerge initially but become less relevant after success has been achieved. From then on, the pool of creative ideas seems to decrease. These views are somewhat supported by the evidence that success may be augmented by pairing high conventionality with novelty using atypical combinations (Uzzi et al. 2013) that themselves may be encouraged by novel interactions. It is also possible that highly innovative researchers such as Nobel laureates may be more critical of new collaborations and may only agree to those that seem to offer meritorious rigor. Moreover, receiving the Nobel Prize might have changed the perception

92 of the laureates with existing or potential coauthors, and vice versa, hence changing the collaboration patterns and structure (Chan et al. 2015). On the other hand, collaborations may be chosen for reasons other than their effect on output or may, over time, transform into friendships, which decreases the pressure to collaborate productively (Hollis 2001). That is, cooperation can lead to an intellectual companionship that overcomes isolation, creating a personal relationship between the coauthors (Katz and Martin 1997). Thus, whereas a new collaboration can enhance diversity of perspective, a long- lasting collaboration may reduce diversity not only in perspective but also in expertise and experience.

It is also important to consider the type of research and the environment in which research is being produced. In-depth research of highly complicated topics requires the assembly of large research teams and may involve very high monetary costs due to specialised and highly technical equipment requirements. It is reasonable to expect that such highly complicated research will yield a continuous stream of data and several layers of complicated yet innovative and important results. Publication of such rich material may lead to a longer lifespan (in terms of publication count) of collaboration between researchers in that research team. Hence the collaboration lifespan depends on the complexity of research topic, however a reduction in diversity of ideas and creative perspective within the same research team is apparently no exception in this case as well.

Conclusion

One definite strength of new collaborations is that these are often characterised by a willingness to consider new ideas and/or adapt to novel approaches. In any collaboration—but particularly in science—trust is crucial to the sharing of ideas, models, data, or material of substantial scientific merit; and the scientific colleagues of Nobel laureates may be more willing to trust Nobelists in that regard. Hence, to benefit from the increasingly collaborative nature of scientific inquiry, researchers need a better understanding of what determines team success. The results reported here suggest that the advantages and costs of ongoing collaboration should be carefully weighed because, from a creativity viewpoint, collaborations have an expiration date, even for Nobel laureates.

Nobel laureates can be seen as (and they probably really are) researchers with evergreen research agenda and research ideas, and yet the impact of their collaboration with the same coauthor diminishes over the lifespan of such collaboration. This is an important lesson for all researchers: one should not underestimate the diminishing returns to collaboration due to stagnation and exhaustion. A crucial strategy for keeping one’s research agenda evergreen is to keep one’s coauthor pool evergreen.

Appendix

Figure 5.4 Distribution of the total number of Nobel laureate coauthors

Note: Bin width = 50.

Figure 5.5 Intensity of cooperation by field weighted by unequal co-author contribution

Figure 5.6 Intensity of cooperation by field weighted by equal co-author contribution

Figure 5.7 Citations received by early and late collaborations of laureate- coauthor pairs weighted by unequal co-author contribution

Figure 5.8 Citations received by early and late collaborations of laureate- coauthor pairs weighted by equal co-author contribution

Figure 5.9 Citations received by first and second half of all collaborations of laureate-coauthor pairs

Table 5.3 A-index for unequal co-author contributions

Coauthors’ share Author position Number of 1 2 3 4 5 6 7 8 9 10 authors 1 1.000 2 0.500 0.500 3 0.417 0.167 0.417 4 0.361 0.194 0.083 0.361 5 0.321 0.196 0.113 0.050 0.321 6 0.290 0.190 0.123 0.073 0.033 0.290 7 0.265 0.182 0.127 0.085 0.052 0.024 0.265 8 0.245 0.174 0.126 0.091 0.062 0.038 0.018 0.245 9 0.229 0.166 0.124 0.093 0.068 0.047 0.030 0.014 0.229 10 0.214 0.159 0.122 0.094 0.072 0.053 0.037 0.023 0.011 0.214

Table 5.4 First Stage Regression Results for 2SLS

Overall Age of the Laureate Chemistry Physics Phy./Med <50 >=50 Collaborations 6-20 0.15*** 0.82*** -0.01 0.26*** 0.29*** 0.50*** (0.055) (0.138) (0.060) (0.098) (0.055) (0.128) Collaborations 21- 0.30*** 1.23*** 0.16** 0.72*** 0.63*** 0.11 40 (0.065) (0.195) (0.070) (0.192) (0.054) (0.198) Collaborations 41- 0.34*** 0.93*** 0.28*** 0.89*** 0.62*** 0.58** 70 (0.075) (0.178) (0.082) (0.225) (0.056) (0.277) Collaborations 71- 0.55*** 1.34*** 0.45*** 0.67*** 0.62*** 2.15*** 110 (0.089) (0.208) (0.100) (0.257) (0.059) (0.359) Collaborations >110 0.74*** 1.49*** 0.65*** 1.23*** 0.74*** 1.16*** (0.120) (0.301) (0.132) (0.245) (0.066) (0.319) Collaboration Year -0.17*** -0.20*** -0.17*** -0.21*** -0.12*** -0.28*** (0.013) (0.031) (0.014) (0.029) (0.011) (0.034) Collaboration 0.01*** 0.01*** 0.01*** 0.01*** 0.01*** 0.01*** Year^2 (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) Number of Authors 0.01*** 0.02*** 0.01*** 0.03** 0.01*** 0.10*** (0.000) (0.002) (0.000) (0.012) (0.000) (0.003) Laureate Fixed Yes Yes Yes Yes Yes Yes Effects Time Fixed Effects Yes Yes Yes Yes Yes Yes Observations 155,179 27,961 127,218 32,473 77,016 45,690 Adjusted R-squared 0.248 0.390 0.184 0.121 0.303 0.195 Notes: First stage coefficients are being reported, where dependent variable is the impact factor of the journal where the article is published. Robust standard errors in parentheses. *p<0.1, **p<0.05, and ***p<0.01.

100

Table 5.5 Descriptive statistics of dependent and independent variables employed in 2SLS regression analysis

Mean Std. Dev. Min Max Citations Received 79.47 395.41 0 8947 Journal Quality (Impact Factor) 7.19 7.56 0.03 51.66 Collaborations 6-20 0.27 0.44 0 1 Collaborations 21-40 0.15 0.36 0 1 Collaborations 41-70 0.11 0.31 0 1 Collaborations 71-110 0.07 0.25 0 1 Collaborations >110 0.05 0.21 0 1 Collaboration Year 3.57 3.75 1 36 Number of Authors 50.71 64.70 2 181

101

Chapter 6 External Influence as an Indicator of Scholarly Importance

Chan Ho Fai, Bruno Frey, Jana Gallus, Markus Schaffner, Benno Torgler & Stephen Whyte

CESifo Economic Studies (2016), 62(1), 170-195.

Abstract

Although the external influence of scholars has usually been approximated by publication and citation count, the array of scholarly activities is far more extensive. Today, new technologies, in particular Internet search engines, allow more accurate measurement of scholars’ influence on societal discourse. Hence, in this article, we analyse the relation between the internal and external influence of 723 top economists using the number of pages indexed by Google and Bing as a measure of external influence. We not only identify a small association between these scholars’ internal and external influence but also a correlation between internal influence, as captured by receipt of such major academic awards as the Nobel Prize and John Bates Clark Medal, and the external prominence of the top 100 researchers.

Introduction

The primary metric currently used to approximate individual scholars’ influence is the number of publications and, perhaps more appropriately, the number of citations in academic journals. This approach also dominates the rankings of both individual researchers and departments and universities. Yet the scholarly activity this metric captures is only narrowly defined and its impact concentrated ‘within’ academia. For example, collections of general academic publications and citations cover a restricted set of publication outlets (excluding books, pamphlets, reports, and newspapers). Likewise, appointment as a university researcher often depends only on the number of publications in ‘top-tier’ academic journals.

103

Meanwhile, it is generally acknowledged that a scholar’s responsibilities and functions span a far broader array of activities, which can be categorized under four rubrics: (1) scholarly publication including activities as referee, editor, or board member, (2) teaching, (3) academic self- governance (for example, serving as department head or dean), and (4) influence on the broader society. Whereas two of these four activity types— scholarly publication and academic self-governance—are ‘internal’ to the university community, influence on the broader society is ‘external’. Teaching is a mix of the two: although an internal activity, it has an external impact via the influence exerted by students after graduation.

In fact, many scholarly institutions have an explicitly stated goal of participating in the wider societal discourse, although this goal differs between disciplines and subdisciplines. For instance, although the general public does not usually expect theoretical physicists to directly impact society, it does expect applied physicists to make a contribution. To economists, the public even ascribes the ability to predict the future course of the economy, which explains the current backlash against the economics profession and its inability to predict the latest financial and economic crisis.

Moreover, although it is crucial to understand how internal activities within academia relate to the outside world, particularly to the economy (Frey, 2006), views on this matter diverge strongly. For example, Clower (1993, p. 23), a former editor of the American Economic Review (AER), claims that ‘[m]uch of economics is so far removed from anything that remotely resembles the real world that it is often difficult for economists to take their subject seriously’. Blaug (1997, p. 3) advances a more devastating verdict: ‘Modern economics is sick; economics has increasingly become an intellectual game played for its own sake and not for its practical consequences’. Even Nobel Prize recipients in economics, such as Leontief (1971), Coase (1994), and Buchanan (2000) have criticised their field for its lack of involvement in real-life issues. Others, although still convinced that economists do have an effect on society, doubt that this influence is beneficial (for example, Galbraith 1975). Long before the latest travails, Friedman (1972, p. 12) admonished, ‘we economists in recent years have done vast harm—to society at large and to our profession in particular— by claiming more than we can deliver’. As evidenced by press coverage in eminent

104 economics magazines (for example, The Economist 1997, 2000), these negative perceptions are also shared outside of academia.

In contrast to this pessimistic view, other economists tend to embrace Keynes’ (1936) famous claim that ‘the ideas of economists and political philosophers […] are more powerful than is commonly understood. Indeed the world is ruled by little else’ (p. 383). Even his intellectual opponent Hayek (1991) posited that ‘economists have this great influence only in the long run and indirectly’ (p. 37). More recently, similar views of economics’ considerable impact on society have been put forward by Dasgupta (1998), Baumol (2000), and Summers (2000), among others. Baumol (2000), for instance, claims that ‘[in economics], the century has been full of accomplishments. New ideas, new directions, and powerful new tools have emerged in the profession. Evidently, our field of study is alive and well’ (p. 38).

It is difficult, perhaps impossible, to empirically analyse the extent to which these strongly contrasting views on the societal influence of the economics profession apply, not least because there exists no single ‘economic view’ that could be acted upon (Frey 2006). Indeed, economists even struggle to find a consensus about what constitutes ‘economics’ (see Brittan 1973; Kearl et al. 1979; Samuels 1980; Frey et al. 1984; Alston et al. 1992; van Dalen and Klamer 1997; Machin and Oswald 1999). The positions upon which they do agree, however, are viewed with scepticism by the wider public. For example, a recent study by Sapienza and Zingales (2013) identifies a considerable gap (of 35 percentage points) between the answers to policy questions given by economists versus average Americans. Interestingly, this gap is largest for questions on which economists agree the most and on which there is the most literature.

Identifying a unidirectional impact of economics on society is further complicated by reverse causation by which society determines the subject matter of economics. The impact of economists is also driven by the demand side. Academic economists, as well as those working for central banks and other financial institutions, can be asked for their advice in the form of commissioned reports and official statements (a practice sometimes criticised as a way of legitimising already decided policies). We therefore approach the issue of social influence by examining the ‘relation between academic

105

economists’ internal and external influences’; specifically, the extent to which the importance ascribed to economists ‘within’ academia (based on publication and citation count24) is reflected in their ‘external’ influence, as proxied by mentions on Internet search engines (particularly Google and Bing).

The remainder of the article is arranged as follows. Section 2 puts our approach into perspective, summarising the various ways in which a scholar’s external influence may be captured. Section 3 describes our measure for external influence, which is based on the number of pages indexed on Google and Bing. Section 4 then reports our results, and Section 5 concludes the article.

Capturing the external influence of economics and economists

To our knowledge, there are no systematic empirical studies comparing academic economists’ internal and external rankings except for one study that measures the external influence of management scholars in the USA (Aguinis et al. 2012). These authors find that a scholar’s standing within the community of management scholars (as measured by citations in academic publications) deviates significantly and often to a high degree from the attention received outside academia (as measured by web pages on Google). It is therefore unwarranted to assume that a researcher well-known in academia is also recognised outside academia and vice versa: some management scholars prominent outside academia (among them best-selling authors) are rarely if ever cited in academic publications.

Moreover, despite a wealth of literature on the possibilities and pitfalls of measuring research quantity and quality based on publications and citations (for example, Cole and Cole 1971; Lindsey 1980; van Dalen and Klamer 2005; Coupé et al. 2010; Arrow et al. 2011; Johnston et al. 2013; Torgler and Piatti 2013), there is little research addressing the public attention

24 One problem with this measure is that some methods/theories become standard so authors outside this paradigm are no longer cited. We are thankful to one of the referees for mentioning this point.

106 received by economics and economists, probably because such attention is difficult to identify and measure. Related discussions are largely descriptive rather than empirical, a problem that we attempt to remedy by distinguishing three different categories of processes that reflect the public influence of economists and economics.

Reflections of the influence via markets Patents and copyrights

In some disciplines, mostly the natural sciences, a scholar’s contributions to society are at least partly captured by the number of patents received and the income they produce. However, despite a great deal of literature on this measure’s adequacy and the many pitfalls involved (for example, Trajtenberg 1990; Hall et al. 2001), the fact that patents play practically no role in economics eliminates them as a possible measure of outside impact for that discipline. For economics, copyrights are more relevant because they refer to both books and articles in scientific journals and other outlets. Yet to our knowledge, no consistent data exist on this topic. Moreover, copyrights may be considered more as an aid to producing and propagating economic ideas rather than as an indicator of the extent of influence wielded. Writing a bestseller and receiving a high copyright income does not necessarily mean that the respective economist’s ideas have great influence.

Speaking fees

Although scholars well-known to the public may demand higher monetary compensation for giving talks outside academia (Chan et al. 2014b), such activity, albeit potentially influential, may serve primarily as entertainment for a select and private group (for example, at company events), with little wider social impact. In addition, systematic data on such remuneration across countries is limited (Hosp and Schweinsberg 2006).

Advisory activities

One potentially useful indicator of economists’ importance is the positions attained and income received by individuals appointed to expert panels. These positions range from membership of a high-level economic advisory board (for example, the Council of Economic Advisors in the USA or the Sachverständigenrat in Germany) to assuming advisory roles in ministries,

107 non-governmental organizations, and companies. Yet, even though comprehensive statistics on such activities may exist for certain areas, and perhaps even countries, there is no database that would allow us to draw meaningful international comparisons.

Reflections of the influence via persons Former students in the private and not-for-profit sectors

Economic knowledge may be transferred to the public by former students who have become active outside academia; for instance, as managers in private firms, as members of interest groups, or as participants in the voluntary sector. This type of influence is difficult to capture because the underlying economic ideas are not necessarily expressed explicitly but rather may have been integrated into the alumni’s thinking and actions. Hence, although students may inculcate the economic ideas and further propagate them, this influence is difficult or even impossible to capture statistically. Admittedly, business school evaluations do try to capture alumni’s potential influence by measuring their subsequent income, yet usually only the starting salary is taken as an indicator of the value added to a person’s educational capital. Obviously, this measure is incomplete and biased, particularly given the significant differences in ‘average’ salaries across different economic sectors. For instance, the salary of a graduate working in the financial sector tends to be much higher than that of a comparable graduate working in the non-profit sector. In an effort to develop a more useful ranking matrix, Research Papers in Economics (RePEc, see http://repec.org) has recently introduced the publication Genealogy, which allows individuals to provide information about their students and supervisors with the aim of assessing dissertation advisor and doctoral program quality. A recent poll by RePEc indicates that 54% of respondents are in favour of such a ranking (RePEc 2013a).

Politicians and public officials

When we limit our attention to economics professors during recent years, we can identify several economists who have achieved high ranks in politics and public administration. In the Netherlands, for example, Lubbers, Zijlstra, and De Quay were all prime ministers; Andriessen, Duisenberg, Witteveen, and

108

Zahn were ministers of finance; and Pronk and Ritzen served as ministers in other departments. In Germany, Erhard was Chancellor; Schiller was finance minister, and Töpfer and Hankel were heads of other ministries. In Italy, Prodi and Monti were both Prime Minister and Einaudi was President of the Republic. In many countries, the position of the president of the central bank is normally occupied by a former professor of economics. We are, however, unaware of any reliable and encompassing data on such positions.

Reflections of the influence via outside markets References in official documents

Official documents offer two potentially effective measures of the extent to which researchers’ contributions have actual policy implications: the first is the citation count in publications released by public bureaucracies; the second is the citation count in commissioned reports and similar materials.

Surveys

A scholar’s importance outside academia may also be captured by surveying the general public (for example, in popular journals) or specific groups, such as public bureaucracies, special interest groups, and not-for-profit institutions.

Awards

Scholars may also receive orders, medals, crosses, prizes, and other awards from institutions outside academia, as typified by the British Queen’s appointment of scientists to the House of Lords. Because such honours signal the importance and quality of the recipient’s work (Frey and Gallus 2014), we examine the relation between external influence and such key awards in economics by analysing data on the Nobel Prize, the John Bates Clark Medal, and various fellowships (Fellow of the Econometric Society, the American Economic Association, or the European Economic Association; see also Chan et al. 2014a).

Publications and citations in the popular media

Members of academia may actively influence society by writing in newspapers or other press venues accessed by the public, including appearances on radio and television. Scholars may also passively influence

109

the wider public via these channels if the topic’s societal interest causes journalists to report on it.

New media

Scholarly activity by economists is also reflected via new media channels, including digitised books and newspaper articles, published mostly online, that are either written by scholars or cite their findings. Beyond these traditional publication outlets on the Internet, influence may also be exerted via Twitter, online portals on which economists present brief but socially relevant economic analyses (for example, VoxEU), and economists’ blogs (for example, ‘Marginal Revolution’), which are sometimes written in collaboration with non-academics (for example, ‘Freakonomics’). According to the Wall Street Journal, the most popular of these blogs can attract as many as 50,000–100,000 page views a day (Evans 2009). These new media, more than any other platform, are interactive and largely unregulated, meaning that persons from outside academia may engage in or launch discussions with economists. Users may thus multiply the reach of economic ideas by sharing and citing them within their social networks, for instance, on Twitter and Facebook or on their own blogs. Interestingly, according to the RePEc poll, 73% of respondents argued against counting Wikipedia mentions as citations for ranking purposes, and 84% were against doing so for blogs (RePEc 2013b).

Methodological approach

We gauge economists’ influence outside academia by using web page counts from the widely used search engine Google (comScore 2014), focusing on 25 pages outside the ‘.edu’ domain (see also Aguinis et al. 2012). These counts reflect how much attention has been paid to a particular economist online; for example, on mainstream media sites, blogs, and social media. To ascertain the reliability of these counts, we also employ counts from the Microsoft search engine Bing. These are of course not the only possible measures of scholars’ societal impact; they do, however, go much further than the citation

25 The sequence .ac is also used in some countries as a second-level domain for academic institutions. At the time of data collection, however, we did not exclude pages from this domain.

110 and publication count measure usually used to assess scholarly influence. Nevertheless, although our measures approximate influence outside of academia, like citation counts, they fail to capture why the author was mentioned (for example, in criticism or praise of his/her work). Citations may be increased, for instance, by negligible mentions in footnotes. Similarly, web page counts may be raised if a scholar has established a dominant online presence (for example, through a blog and Twitter usage), or these counts may reflect media firestorms provoked by the work or an interview given in the press (The Economist 2015). Hence, like the citation count, the web page count is no measure of the quality of an author’s work.

Our initial sample of academic economists was drawn from the September 2012 rankings in RePEc/IDEAS, the largest freely available bibliographic database on the Internet dedicated to economics and finance (http://ideas.repec.org). RePEc covers more than 43,000 registered academic researchers who are evaluated monthly on a range of publishing measures (as of February 2015). We use RePEc’s average rank score (which takes the harmonic mean of various rankings) to select the top 1000 researchers (http://ideas.repec.org/top/top.person.all.html), a method that mirrors Aguinis et al.’s (2012) use of current webometric techniques to explore the impact of the top 550 management scholars. By doing so, however, we are exploring relatively successful scholars who are not representative of the entire academic population of economists.

After first conducting searches using quotation marks around author names to avoid spurious matches (and thus incorrect crediting of webpage counts), we controlled for the validity of the sum of each individual’s counts by running a single search in two versions of Google (the American google.com and the Swedish google.se). The total number of pages was identical for both versions, a consistency also reported in Aguinis et al.’s (2012) comparison of the American and Spanish versions. Next, to deal with any spurious matches generated by results that were clearly unrelated to the author in question, we employed Aguinis et al.’s (2012) criterion of 5% spurious entries to exclude authors and increase the integrity of the data set by avoiding possible upward bias in the total number of web pages. That is, for all 1000 researchers, we manually checked the first 50 pages, and if three or more pages were not attributed to the author, we excluded this individual from the sample. Finally, to alleviate any concerns about fluctuations in the

111

count totals for Google pages, we ran four separate searches across an 18-day period (19 October to 6 November), with the first collection conducted manually by three research assistants over a 24-hour period on 10 October.

From this initial manual search, we eliminated 274 of the 1000 authors based on the 5% criterion for non-attributed pages. During this manual collection, we also identified discrepancies between the spelling of an author name in RePEc (used purely to define and classify registered authors) and the actual name used in publications, thereby accounting for the fact that many academics publish under more informal or more socially well-known names (for example, ‘Mark L. Gertler’ in RePEc appears as ‘Mark Gertler’ on all his published work). In total, we identified 69 names with inconsistencies such as multiple middle names and extra or removed middle initials. Because any Google search for two different names (even for the same author) would result in a different page count, we excluded these cases from our data set, leaving us with a final sample of 723 researchers.

To extrapolate and aggregate the page counts for each individual, we relied on three automated computer searches taken directly from the Google and Bing Search application programming interfaces (APIs), which allow a large number of searches to be run simultaneously in a short amount of time. These three ‘automated’ searches (run on 31 October, 2 November, and 6 November) not only ensured more robust data capture but also reduced the potential for human error.

Although the number of total search results reported in the manual and automatic counts are only estimates (process not disclosed), the API searches produced a significantly lower estimate than the manual searches. We can only speculate that the results for the manual searches could be slightly inflated as a result of the search engine’s extensive index, while the automatic search results reflect an underestimate based on the preliminary search. Hence, whereas the manual search returned a value of 5,410,000 pages for the researcher with the strongest external impact, the API returned only 922,667 pages (over an average of three search processes). Nevertheless, both values are highly correlated.

To avoid limiting the search scope and to further the argument for capturing a wider social impact, we conducted all Google automated searches concurrently on the secondary search engine, Bing. The very high scale

112 reliability coefficients (Cronbach’s alpha) for our different count days (Google = 0.9998, Bing = 0.9812) are comparable to those achieved by Aguinis et al. (2012) and justify computing an average based on the total number of Google or Bing entries across the three automatic data collection waves. The average Google and Bing entries are strongly correlated (Pearson’s r=0.829) at a 1% level of significance.

To construct a proxy for the impact ‘inside’ academia for the 723 economists, we take the individual rankings provided by RePEc based on three baseline measures: total number of citations, total number of articles (and pages), and the h-index (which assigns, for example, a score of 30 when 30 of a scholar’s papers have at least 30 citations each but his/ her other papers have no more than 30 citations each). To increase the robustness of the analysis, we also include author rankings based on the weighted values of the citations and articles (by number of authors and the journals’ simple impact factor, see http://ideas.repec.org/top/). Combining these various measures enables fuller evaluation of different aspects of a scholar’s academic performance. For example, the total number of citations for a scholar’s works (with self-citations excluded) captures the total attention by peers, while total number of publications provides a measure of productivity. The h-index provides an intuitive measure that takes into account both citation and 26 publication counts. We also explore academic influence by evaluating academic recognition as reflected by the following awards and honours: the John Bates Clark Medal, the Nobel Prize, the Frisch Medal, Fellow of the Econometric Society, Fellow of the European Economic Association, Distinguished Fellow of the American Economic Association, or Foreign Honorary Member of the American Economic Association.

26 Other indices derived from the h-index are also possible. One alternative would be Wu’s (2010) w-index, which improves on the h-index by placing more focus on the influence of a scholar’s top cited papers. The w-index, however, is less able to differentiate between scholars with fewer citations, thereby producing less variation in the rankings than does the h-index. We therefore did not collect data on the w- index for our analysis. Nevertheless, taking the top 2000 RePEc researchers for both indexes, we observe only 19 different ranking positions for the w-index but 46 for the h-index (as of January 2015) (pp. 12–13).

113

Results and discussion

Table 6.6 reports the 100 most influential economists based on the number of Google pages, together with their RePEc ranking (average rank score) and Bing page values (for comparative purposes). Nobel laureate Milton Friedman, a very active public figure, leads the list; followed by Nouriel Roubini, who was active with the International Monetary Fund, the Federal Reserve, the World Bank and the Bank of Israel; and Nobel laureate Amartya Sen. Also in the top 10 are Nobel laureate Daniel Kahneman, as well as laureate Joseph Stiglitz, who also won the John Bates Clark Medal and served as senior vice president and chief economist of the World Bank and as member and Chairman of the Council of Economic Advisers. In the top 20, we note Alvin Roth, who received the 2012 Nobel Prize in economics.

Several other researchers who combine academic research with policy making are also ranked highly, including Dani Rodrik, who has conducted substantial work on economic policy and government performance, at rank 4; Oliver Blanchard, chief economist at the International Monetary Fund, at rank 8, and Ben Bernanke, chairman of the Federal Reserve, at rank 10. Also on the list at rank 11 is Hans-Werner Sinn, president of the Ifo Institute for Economic Research and since 1989, a member of the Advisory Council of the German Ministry of Economics. Ranked at number 13 is Australian economist John Quiggin, chief research economist with the Bureau of Agricultural Economics, board member of the Climate Change Authority of the Australian Government, and a very active blog writer. The top 20 also includes two other John Bates Clark Medallists (JBCMs), Steven Levitt (at rank 15) and Daron Acemoglu (at rank 17), who are also very successful book authors. Levitt’s Freakonomics and SuperFreakonomics, co- authored with Stephen Dubner, have received wide media and readership attention, leading to a blog, radio show, and movie (see http://www.freakonomics.com/), while Acemoglu’s Economic Origins of Dictatorship and Democracy, co-authored with James Robinson, accounts for more than 2300 Google Scholar citations (as of 28 July 2013). Ranked at number 18 is Andrei Shleifer, also a JBCM and a key figure in the Russian privatization process, who leads the RePEc ranking. Rounding out the list at number 20 is William Easterly, who has worked for 16 years as a researcher at the World Bank, published widely read books (for example, The Elusive

114

Quest for Growth), and maintained a major public debate on foreign aid with his adversary Jeffrey Sachs.

Figure 6.1 presents two Lorenz curves that illustrate the inequality in the societal influence of individual economists (as measured by Google and Bing counts). In line with Aguinis et al.’s (2012) findings, the figure shows that individual performance follows a power law distribution. A minority of economists draws the majority of web page counts: 20% of those listed are responsible for around 70% of the Google page counts. The Gini coefficient is similarly large (0.690 compared with 0.433) when Google is used instead 27 of Bing. Such highly skewed distributions have been observed in many fields, ranging from biology to social networks (Simon 1955; Barabási and Albert 1999; Barabási 2003). A related concept, the winner-take-all principle (Rosen 1981; Frank and Cook 1995), suggests that minimal differences in individual performance are enough to generate such huge outcome differentials.

Figure 6.1 Lorenz curves for Google and Bing page counts

Next, we analyse the discrepancy between the rank ordering of economists produced by our measure of societal influence and that produced by the standard RePEc measure. The average rank order difference (ranking of Google page counts minus RePEc average ranking) amounts to 175.5

27 The Gini coefficient of the number of Google pages (non-.edu domains) in Aguinis et al. (2012) is also 0.688.

115

ranks, which is a considerable change and conforms to the difference of on average 100 ranks found by Aguinis et al. (2012) in their comparison of the 28 lists produced by citation counts and Google entries. The two-sample Kolmogorov–Smirnov test also suggests that the distribution of the rank order difference in our sample is not significantly different to that in Aguinis et al. (2012) (see Figure 6.5). Figure 6.2 illustrates the frequency of the different ranking discrepancies that arise when switching between the RePEc and Google (or Bing) measures. In our data set, there is a difference between the Google (Bing) and RePEc listings of over 150 ranks for 47.2% (56.8%) of all scholars. Timothy Besley, for example, is ranked at 67 in RePEc but only at 620 based on Google counts. Similarly, Eugene Fama has a RePEc ranking of 49 but a Google ranking of only 712. Conversely, Simon Kuznets has a RePEc ranking of 678 but a Google ranking of 47, and Reinhard Selten ranks at 585 in RePEc but as high as 75 on Google counts. Likewise, Raj Chetty (who recently received the John Bates Clark Medal) is ranked at 631 on RePEc but at 194 on Google counts. Overall, the histograms in Figure 6.2 show quite clearly that a scholar’s ranking position among fellow economists can differ dramatically based on whether the ranking is based on internal academic evaluations or societal impact. Tellingly, Aguinis et al. (2012) cite an anonymous reviewer in the Academy of Management Perspectives who claims that the results obtained ‘should give administrators pause’ (p. 115).

28 The initial RePEc ranking is reordered for comparison to the rank order by Google and Bing page count so that all three rankings take values from 1 (highest) to 723 (lowest).

116

Figure 6.2 Rank order differences across lists, by either internal or external influence measure

117

To further analyse the relation between external impact and other possible measures of internal academic performance, the first two columns in Table 6.1 present an overview of the correlations between web page counts and a large set of academic performance metrics taken from RePEc for the sample of 723 economists. Here, the correlation measure is Kendall’s tau rank correlation (s) rather than the standard Pearson product-moment correlation (r) because the RePEc rankings are ordinal variables whereas the number of web entries is a continuous variable. Unlike Pearson’s r, the Kendall tau rank correlation does not require linear statistical dependence between two variables, only monotonicity. Specifically, it measures the difference between the proportion of the number of concordant pairs (for example, the larger of the two RePEc rankings is associated with the larger of the two Google or Bing rankings) and the number of discordant pairs (for example, the larger RePEc rank order is allied with a smaller Google or Bing ranking) by the total number of pair combinations.

For both Google and Bing, all academic performance ranking metrics 29 are significantly and positively correlated with external influence. Among all internal influence measures, the rankings on ‘average rank score’ have the highest positive correlation with the number of Google pages (0.277) followed by ‘number of citations’ (0.254) and ‘h-index’ (0.252). In comparison, the correlation in Aguinis et al. (2012) between Google entries 30 and the number of citations by management researchers equals 0.237. The Bing metric overall produces a weaker positive correlation with the RePEc rankings. For example, the correlation between external influence and the RePEc ‘average rank score’ ranking equals 0.132, while the correlations for ‘number of citations’ and ‘h-index’ equal 0.098 and 0.102, respectively. It therefore seems that internal academic impact is only weakly correlated with external impact, meaning that the importance ascribed to an economist within academia is only partially reflected in the scholar’s external influence.

29 The level of significance of the Spearman’s rank correlation does not differ from the Kendall’s tau correlations. 30 Aguinis et al. (2012) report a Pearson’s correlation of 0.166 for total number of citations and total number of non-.edu Google pages. To enable comparison, we calculate the Kendall’s tau rank correlation using data from Table 4 in Aguinis et al. (2012).

119

Table 6.1 Correlations between external influence and internal performance (RePEc rankings)

RePEc rankings Average: Google (N=723) Average: Bing (N=723) Average: Google (Top 100) Average: Bing (Top 100) 0.2771*** 0.1324*** 0.2847*** 0.0752 Average rank score (0.0000) (0.0000) (0.0000) (0.2692) 0.1271*** 0.0972*** 0.0265 -0.0188 Number of distinct works (0.0000) (0.0001) (0.6986) (0.7841) Number of distinct works weighted by simple impact 0.2211*** 0.1074*** 0.1113 -0.0012 factor (0.0000) (0.0000) (0.1014) (0.9881) Number of distinct works weighted by number of 0.2186*** 0.1430*** 0.0879 0.0230 authors and simple impact factor (0.0000) (0.0000) (0.1962) (0.7365) 0.1283*** 0.0774*** 0.0004 -0.0697 Number of journal pages (0.0000) (0.0019) (0.9976) (0.3056) Number of journal pages weighted by simple impact 0.1897*** 0.1050*** 0.1457** 0.0473 factor (0.0000) (0.0000) (0.0320) (0.4877) Number of journal pages weighted by number of 0.1842*** 0.1299*** 0.1445** 0.0942 authors and simple impact factor (0.0000) (0.0000) (0.0335) (0.1661) 0.2536*** 0.0984*** 0.2698*** 0.0772 Number of citations (0.0000) (0.0001) (0.0001) (0.2565) 0.2222*** 0.0775*** 0.3150*** 0.1152* Number of citations weighted by simple impact factor (0.0000) (0.0018) (0.0000) (0.0902) Number of citations weighted by number of authors 0.2294*** 0.1069*** 0.2702*** 0.1536** and simple impact factor (0.0000) (0.0000) (0.0001) (0.0238) 0.2519*** 0.1017*** 0.2758*** 0.0684 h-index (0.0000) (0.0001) (0.0001) (0.3235) Notes: The table reports Kendall's Tau-b (ties adjusted) statistics, with p-values in parentheses. For ease of interpretation, in all correlations, all the RePEc ranking values are multiplied by -1 so that larger values represent higher ranks. Therefore, a positive correlation means that higher academic performance is associated with higher external influence. *, **, and *** represent statistical significance at the 10%, 5%, and 1% levels, respectively.

121

Admittedly, it could be argued that exploring only the top 100 academics (according to RePEc ranking) could produce different outcomes than this general sample of 723 economists. Yet according to the two columns on the right-hand side of Table 6.1, such is not the case: when the sample is much smaller, the correlations become less significant overall. Compared to previous findings, in which the Bing web page count produced various significant correlations with academic performance proxies, only ‘number of citations weighted by simple impact factor’ and ‘number of citations weighted by number of authors and simple impact factor’ show a statistically significant correlation (10 and 5%, respectively). For the Google pages, external influence is positively correlated with rankings based on ‘average rank score’, citation measures, and ‘h-index’ at a 1% level of significance and with ‘number of journal pages weighted by simple impact factor’ and ‘number of journal pages weighted by number of authors and simple impact factor’ at a 5% level.

The RePEc rankings, however, can be criticised as lower-bound performance measures because rather than registering all the economics journals, the RePEc generates citations by extracting the list of references (http://citec.repec.org/) from each document made available to its digital library in electronic format. At present, because of software limitations in reference identification (that is, PDFs must be converted to ASCII) and related requirements that the documents must satisfy (http://citec.repec.org/warning.html), only around 74% of these records have been analysed. Hence, in Table 6.2, we also employ metrics from two other sources used in economics and beyond, Publish or Perish (version 3), which enables a wide range of publishing metrics (see also Harzing 2010), and the Web of Knowledge.

Because many authors publish across different disciplines, we conducted both these searches with no constraints on journal of publication, which allowed us to capture the total internal academic impact rather than simply the specific impact on the author’s primary research field. The searches were conducted within a 72-hour period (from 1 to 3 March 2013) to ensure as little variation as possible over time. As in the two columns on the right-hand side of Table 6.1, we restrict our analysis to the top 100 economists in the RePEc rankings.

123

As Table 6.2 shows, when using the Publish or Perish data, the different h-index scores show the strongest positive correlations with external impact, ranging from 0.227 (e-index with Google) to 0.495 (hI-index with 31 Bing). Nevertheless, external influence is not correlated with the important success metric, citations per paper; the correlations with age-weighted citation rate are either not statistically significant or have only borderline significance; and the Web of Knowledge metric shows no correlation with external influence for the top 100 researchers. Table 6.2 therefore supports the earlier observation: there is no (or only a weak) correlation between internal success and external influence.

Table 6.2 Correlations between external influence and total internal academic impact (Publish or Perish and Web of Knowledge)

Average: Average: Publish or Perish Google Bing Citations 0.2361** 0.2928*** (0.0181) (0.0031) Citations/years 0.2054** 0.2750*** (0.0403) (0.0056) Citations/papers -0.0125 0.0287 (0.9017) (0.7769) Average N papers per author -0.2187** -0.2375** (0.0288) (0.0174) h-index 0.3005*** 0.3938*** (0.0024) (0.0001) g-index 0.2559** 0.3361*** (0.0102) (0.0006) hc-index (Contemporary h-index) 0.2672*** 0.3305*** (0.0072) (0.0008) hI-index (Individual h-index) 0.4033*** 0.4946*** (0.0000) (0.0000) hm-index (Individual h-index) 0.3760*** 0.4923*** (0.0001) (0.0000) AWCR (Age-weighted citation rate) 0.1178 0.1347 (0.2433) (0.1815) AWCRpA (Normalised to the number of 0.1695* 0.1827* authors) (0.0918) (0.0688) e-index 0.2273** 0.2916*** (0.0230) (0.0032) Web of Knowledge Total citation count 0.0665 0.0365 (0.5112) (0.7184) Average annual citation 0.0545 0.0004 (0.5905) (0.9971)

31 Because all variables in Table 6.2 are continuous variables, we report the Pearson product–moment correlation.

124

Notes: The table reports the Pearson product-moment correlation, with p-values in parentheses; n = 100. The Publish or Perish metrics are described in detail in Harzing (2010) or at http://www.harzing.com/pop.htm. *, **, and *** represent statistical significance at the 10%, 5%, and 1% levels, respectively.

‘Superstardom’, considered above by focusing on the top 100 economists, is also evident in the extensive system of awards on which academia relies and which serves as a tool for distinction. The most renowned award after the Nobel Prize is the John Bates Clark Medal awarded to a scholar under 40 ‘who is judged to have made the most significant contribution to economic thought and knowledge’ (http://www.aeaweb.org/ honors_awards/clark_medal.php). Becoming a Fellow of the Econometric Society is also considered prestigious (Hamermesh and Schmidt 2003) despite the substantial number of fellows (877 by the end of 2011; see Chan and Torgler 2012). Many JBCMs and Economic Society fellows later became Nobel laureates. The other awards for academic economists, although also prestigious, can be classified as somewhat less important. In Table 6.3, we show that overall, the level of the respective award’s prestige is positively and significantly correlated with external influence as measured by the web page counts. The highest correlations are observed for the Nobel Prize (r=0.290) and the John Bates Clark Medal (r=0.249), although external influence is also significantly positively correlated with being a Fellow of the Econometric Society and an Emeritus Fellow of the European Economics Association. In other cases, there is barely any correlation.

Table 6.3 Correlation between external influence and prizes and awards received

Number Average: Average: of Google Bing awardee s JBC Medal 0.2485*** 0.2116*** 18 (0.0000) (0.0000) Nobel Prize 0.2451*** 0.2902*** 37 (0.0000) (0.0000) Frisch Medal -0.0190 -0.0212 13 (0.6109) (0.5699) Distinguished Fellow of the AEA 0.0298 0.0974*** 23 (0.4234) (0.0088) Foreign Honorary AEA -0.0035 0.0008 19 (0.9251) (0.9830)

125

Fellow of the Econometric 0.0972*** 0.1382*** 250 Society (0.0089) (0.0002) Fellow of the EEA 0.0060 0.0286 32 (0.8715) (0.4432) Emeritus Fellow of the EEA 0.1325*** 0.1413*** 23 (0.0004) (0.0001) Notes: The table reports the Pearson product-moment correlation, with p-values in parentheses; n = 723. *, **, and *** represent statistical significance at the 10%, 5%, and 1% levels, respectively.

Nevertheless, it remains unclear whether these results are driven by the likelihood that the Nobel Prize or the John Bates Clark Medal will be given to a scholar with a strong external influence or by the fact that these awards actually have a positive impact on external influence. Figure 6.3 and 32 Figure 6.4 graph the results of a Google trend analysis conducted between 25 and 27 February 2013, which extrapolated relative monthly search volume counts for each JBCM (N=6) and Nobel laureate (N=16) receiving the award between January 2005 and January 2013 (a 97-month period). A massive peak is evident in the month in which the award was announced, indicating a relatively high interest from the general public in the current winner of both awards. The relative search volume count is also statistically higher (at a 1% level of significance using a mean comparison t-test) during the 50 months ‘after’ award conferral (excluding the month of award announcement) than 33 during the 50 months ‘before’ it. This observation could suggest that the Nobel Prize and John Bates Clark Medal have a positive impact on external influence.

32 See http://www.google.com/trends/. 33 The difference in search volume remains statistically significant at a 1% level even when we exclude 5 months before and 5 months after the award announcement.

126

Figure 6.3 Google trends for Nobel laureates before and after the Nobel Prize

Figure 6.4 Google trends for John Bates Clark Medallists before and after the award

We then correct for the possibility that external influence might be substantially driven by the social attention following award reception by

127

running Google and Bing searches on whether each scholar is a JBCM or Nobel Prize winner. We subtract the number of Google pages found from a normal name search result to eliminate Google/Bing hits related to the award. The correlation values obtained (see Table 6.7) are higher than those derived earlier for Bing but lower for Google (in particular for Nobel Prize winners). The robustness of the results for the JBCMs could indicate that researchers with a higher external influence are more likely than other researchers to earn this prestigious medal.

To check whether the correlations remain the same once other factors are controlled for, we conduct a multivariate analysis using the ordinary least squares linear regression model. In particular, given the empirical evidence that educational background shapes academic researchers’ career success (Chan and Torgler 2013), we examine how this background affects or even accentuates scholars’ social impact. Again, we narrow our performance criterion to researchers with a strong recent publication performance, defined as at least one publication in the AER, Econometrica, or the Journal of Political Economy between 2005 and 2010. From among the over 1200 academics who published work in these three journals across the 6-year period considered, we identify 193 out of the 723 academics in our revised RePEc top 1000 list. We use the curriculum vitae of each academic to identify their doctoral university and year of graduation, and thus their academic age. To measure their university ranking position, we use the classification developed by Amir and Knauff (2008), which ranks the top 58 economics universities globally based not on research productivity but on the strength of the Ph.D. program as measured by the department’s ability to place doctoral graduates in top-level economics departments or business schools. Because the ranking goes from 1 to 58, we classify all the universities with a constant value of 59, allowing us to create Top 10 and Top 20 dummies.

Based on the above variables, we then consider the following model:

= + + + +

𝑊𝑊𝑊𝑊𝑊𝑊𝑊𝑊𝑊𝑊𝑊𝑊𝐸𝐸𝑖𝑖 𝛽𝛽0 𝛽𝛽_1𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐷𝐷+𝑖𝑖 𝛽𝛽2𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝐶𝐶𝑅𝑅𝑅𝑅𝑅𝑅_ 𝐾𝐾i 𝛽𝛽+3𝑀𝑀𝑀𝑀 𝑀𝑀𝐸𝐸𝑖𝑖 4 𝑖𝑖 5 𝑖𝑖 𝑖𝑖 where i indexes𝛽𝛽 the𝐼𝐼𝐼𝐼𝐼𝐼 economists𝑅𝑅𝑅𝑅𝑅𝑅𝐾𝐾 𝛽𝛽 in𝐴𝐴 𝐴𝐴the𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴 sample,𝐴𝐴 𝐴𝐴WEBPAGE𝐸𝐸 𝜀𝜀 i denotes the

economist’s number of non-.edu Google of Bing pages, AWARDi is a dummy

variable for JBCM or Nobel laureate, REPEC RANKi is the RePEc ranking

base on the ‘average rank score’, MALEi is a dummy variable for the scholar

128 being a male, INS RANKi is the institutional ranking (or dummy variables for top 10 and top 20 universities), and ACADEMIC AGEi is the number of year since doctoral degree graduation. εi denotes the error term. The first two specifications (1a, 1b) in Table 6.4 are based on the 1–59 institutional ranking and include the dummy for whether a scholar is a JBCM or Nobel laureate. The next four specifications contain the dummies for Top 10 (2a, 2b) and Top 20 institutions (3a, 3b). Specifications (4a) and (4b) then further differentiate between JBCMs who are not Nobel laureates, Nobel laureates who are not JBCMs, and JBCMs who are also Nobel laureates.

As the results clearly show, ceteris paribus, the recipients of these prestigious awards generate substantially more external influence (for example, 17,800 more Google web pages) than all the other top researchers. The academics with the strongest performance are those who earned both the John Bates Clark Medal and the Nobel Prize. Again, however, the RePEc ranking is statistically significant only in the Google search process, never in the Bing analysis. This observation is confirmed in Table 6.5 by the fact that none of the sub-factor ranking variables are statistically significant in the Bing regressions. Interestingly, males seem to generate more external influence than females, but the institutional ranking of the doctoral university has no influence on external impact. In addition, when the Google pages are used as the dependent variable, our evaluation of a scholar’s influence reveals a negative relation between external influence and academic age (years since Ph.D. or highest education). This finding suggests that a less senior economics scholar may be able to counterbalance shortcomings such as fewer citations by exerting an important impact outside academia.

The analysis in Table 6.5 examines the major sub-factors reported in Table 6.1 in place of the overall ranking information. As the table shows, these sub-factors, like the overall rankings, are not statistically significant. When Google pages are the dependent variable, however, the citation proxies and weighted journal pages do reach statistical significance. These findings do not change in the robustness tests carried out using log Google and Bing values as the dependent variable: award recipients tend to have more external impact.

129

Table 6.4 Determinants of external influence

Dependent variable Google Bing Google Bing Google Bing Google Bing (1a) (1b) (2a) (2b) (3a) (3b) (4a) (4b) JBCM or Nobelist 17799.8*** 1627.2*** 17924.1*** 1632.1*** 17802.7*** 1633.5*** (3.12) (5.14) (3.16) (5.17) (3.14) (5.16) JBCM but not Nobelist 26751.8*** 955.4* (4.17) (1.90) Nobelist no JBCM 13754.1*** 2120.0*** (2.96) (5.81) JBCM and Nobelist 14267.2* 1116.7* (1.69) (1.69) RePEc ranking 18.5*** 0.3 18.8*** 0.3 18.5*** 0.3 17.4*** 0.4 (4.94) (1.10) (5.05) (1.20) (4.87) (1.18) (4.66) (1.41) Male 6700.1*** 618.6*** 6571.9*** 613.0*** 6690.2*** 618.1*** 6500.8 630.4* (3.75) (2.80) (3.71) (2.77) (3.80) (2.80) (1.44) (1.78) Institutional Ranking 3.1 0.0 (0.08) (0.00) Academic age -272.5*** -11.0 -275.0*** -11.1 -272.9*** -11.0 -205.3* -16.0* (-2.72) (-1.32) (-2.77) (-1.35) (-2.73) (-1.34) (-1.78) (-1.77) Top 10 institution -1007.8 -48.1 (-0.56) (-0.31) Top 20 institution 171.0 -52.2 135.6 -45.7 (0.10) (-0.27) (0.05) (-0.22) N 193 193 193 193 193 193 193 193 R-squared 0.250 0.184 0.251 0.184 0.250 0.184 0.261 0.204 Prob. > F 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 Notes: The values of RePEc ranking and Institutional Ranking are multiplied by -1. t-statistics are in parentheses. *, **, and *** represent statistical significance at the 10%, 5%, and 1% levels, respectively.

131

Our estimates, however, can only provide an overview of the relation between internal impact and external influence because our cross-sectional analysis, unlike a longitudinal analysis, cannot observe the evolvement and dynamics of how internal success impacts the societal influence of an academic (causal relation). The sample selection bias resulting from the exclusion of observations based on the 5% spurious entries rule might also raise concerns, especially when there is limited access to the actual search algorithm employed by Google and Bing.

In general, our methodological approach has two major problems: some search results may have been included that should not have been whereas some authors may not be included that should be, which could result in measurement errors. For example, there might be cases in which the first 50 pages but not the remaining web pages (or vice versa) refer to a scholar that the search algorithm (for example, Google PageRank) ranks as having the highest relevance. This situation would introduce both upward and downward biases when the number of web pages is used to proxy external influence. One possible remedy would be to increase the number of search results assessed manually and look not only at the first 50 pages but also at the middle and last 50 pages.

Similarly, it is debatable how well the Google and Bing web pages actually capture a scholar’s influence in society. Despite little doubt that most traces of a scholar’s external influence can be found in all these web pages, the question remains of how strongly the total number of web pages is correlated with the scholar’s intensity in, for example, giving advice to politicians, contributing to public policy change, or acting in the media. Using the top 100 economists in Table 6.6, we find that the number of Google pages has a strong positive correlation with the number of entries in Google News 34 (r = 0.82, significant at 1% level), indicating a strong correlation between this narrow proxy of external influence and that used in our analysis.

34 Data collected on 11 February 2015.

133

Table 6.5 Determinants of external influence (extended version)

Dependent variable Google Bing Google Bing Google Bing Google Bing (5a) (5b) (6a) (6b) (7a) (7b) (8a) (8b) JBCM but not Nobelist 28970.9** 1081.5** 28589.0** 1048.7* 28392.2** 1011.0* 28598.7** 1040.9* (2.10) (2.00) (2.08) (1.95) (2.09) (1.88) (2.10) (1.95) Nobelist no JBCM 12627.9** 2079.9*** 13129.4** 2097.2*** 13331.4** 2111.9*** 12594.7** 2091.0*** (2.03) (5.08) (2.10) (5.08) (2.12) (5.14) (2.04) (5.17) JBCM and Nobelist 14033.1*** 1123.9*** 13578.6*** 1130.5*** 14496.8*** 1134.6*** 14234.0*** 1130.5*** (3.06) (3.68) (2.88) (3.66) (3.07) (3.62) (3.18) (3.68) RePEc rankings # journal pages 0.4 0.0 (0.36) (-0.60) # citations 8.3*** 0.1 (3.76) (0.53) # journal pages 5.1** 0.0 weighted by simple (1.99) (0.22) impact factor # citations weighted by 6.7*** 0.0 simple impact factor (2.94) (0.30)

135

Table 7.5 continued

Dependent variable Google Bing Google Bing Google Bing Google Bing (5a) (5b) (6a) (6b) (7a) (7b) (8a) (8b) # journal pages 3.8** 0.1 weighted by number (2.13) (0.60) of authors # citations weighted by 7.9*** 0.1 number of authors (3.47) (0.44)

h-index 10.3*** 0.1 (4.08) (0.53) Male 7945.4*** 692.2*** 5849.1*** 629.3*** 5973.4*** 595.9** 6696.4*** 638.8*** (4.38) (3.12) (3.45) (2.73) (3.40) (2.52) (3.47) (2.92) Institutional Ranking 23.5 0.6 26.4 1.1 3.1 0.7 19.8 1.0 (0.56) (0.13) (0.63) (0.27) (0.08) (0.15) (0.50) (0.22) Academic age -101.4 -10.7 -87.9 -12.1 -143.4 -14.0 -101.8 -12.3 (-1.19) (-1.24) (-1.04) (-1.38) (-1.63) (-1.55) (-1.28) (-1.48) N 193 193 193 193 193 193 193 193 R-squared 0.230 0.198 0.231 0.197 0.237 0.199 0.235 0.197 Prob. > F 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 Notes: The values of all RePEc rankings and Institutional Ranking are multiplied by -1. t-statistics are in parentheses. *, **, and *** represent statistical significance at the 10%, 5%, and 1% levels, respectively.

136

Conclusions

Measuring scholarly activity has emerged as an important topic, not least because of university administrators’ strong incentives to find metrics for departmental progress, as well as academics’ keenness to assess their relative professional standing and the quality of their university environment (Scott and Mitias 1996; Torgler and Piatti 2013). To date, however, academia has relied on a narrow set of internal factors such as publications or citations for its national and international comparisons. Yet in reality scholarly impact is multidimensional (Aguinis et al. 2012) and includes a variety of tasks. Academics can, for example, be categorized as either insiders or locals who are strongly involved in institutional services and in close interaction with members of the same university, or as outsiders and cosmopolitans who bring new ideas, research quality, and outside prestige to the university through their research and activities in national and international professional organizations (Klahr 2004; Wilson 2013). Teaching and academic self- governance can also be classified as local activities despite the external influence later exerted by former students. Academic influence on the broader society, however, goes well beyond the local level, especially in the face of new technologies that enable broader measurement of scholars’ influence in the wider societal discourse. Yet little research has been done on such external influence, a void that this article aims to fill by examining how internal measures of influence within academia relate to the external influence of these same scholars.

Our analysis of the number of Google and Bing web page counts of 723 economics scholars reveals only a low correlation between internal and external influence. This result holds even though we employ a large set of metrics for internal influence, namely weighted and unweighted journal publications, citations, and the h-index. We do identify a difference between academic and external rankings of more than 150 positions for almost 50% of the scholars in our data set. However, although our analysis of the top 100 researchers in RePEc shows a small association between external influence and our academic performance variables, our alternative data source, the Web of Knowledge, reveals no significant correlations between these two variables. The results for the Publish or Perish data, however, are somewhat more positive (correlations up to 0.5). Taken together, these results support Aguinis et al.’s (2012) findings for scholars in management: their impact

137 within academia is not mirrored in their external influence. Rather, our examination of the impact of academic economists suggests that external influence is more strongly correlated with the reception of major awards like the John Bates Clark Medal and Nobel Prize.

Our findings raise many questions for future investigation, including how and why scholars achieve high levels of external influence. One important goal might be to explore fluctuations over time, an approach made possible by the Google and Bing search engines. Also worth considering are other sources that capture external influence; for example, new and popular media, official documents, and patents.

138

Appendix

Figure 6.5 Rank order differences compared to Aguinis et al. (2012)

Note: The Aguinis et al. data are taken from Table 6.4 (pp.116–124). The p-values correspond to the two sample Kolmogorov-Smirnov equality-of-distributions test.

139

Table 6.6 Ranking of economics scholars by average number of web counts

Name Ranking: Ranking: Ranking: Average # Google Bing RePEc of (Average Rank Google Score) pages Milton Friedman 1 2 265 922667 Nouriel Roubini 2 1 596 712000 Amartya Sen 3 5 231 412000 Dani Rodrik 4 3 59 274000 Christopher F. Baum 5 618 15 262000 Daniel Kahneman 6 10 178 229667 Joseph E Stiglitz 7 4 3 222000 Olivier Blanchard 8 19 12 153333 N Gregory Mankiw 9 55 32 122000 Ben S Bernanke 10 11 29 118000 Hans Werner Sinn 11 7 143 111667 Michele Boldrin 12 14 447 105333 John Quiggin 13 9 277 103667 Alvin E Roth 14 17 117 79833 Daron Acemoglu 15 41 6 70567 Austan Goolsbee 16 12 831 68267 Steven Levitt 17 27 189 67633 Andrei Shleifer 18 284 1 65933 Kaushik Basu 19 6 367 62933 William Easterly 20 79 110 61300 Luigi Zingales 21 49 89 58767 Andreu Mas Colell 22 15 678 49067 Paul A Samuelson 23 37 169 48433 Lawrence H. Summers 24 68 23 45800 Robert J. Shiller 25 113 82 44367 Esther Duflo 26 82 227 43733 Lars E. O. Svensson 27 374 48 43700 Jean Tirole 28 264 8 42967 Lucrezia Reichlin 29 16 339 42833 John B Taylor 30 18 56 42833 James Poterba 31 372 41 42033 Jonathan Gruber 32 29 219 39333 Tito Boeri 33 60 836 38767 Franco Modigliani 34 36 610 38700 Xavier Sala Imartin 35 35 152 38633 Gary Gorton 36 91 331 38400 Thomas Piketty 37 31 613 38167 John List 38 23 77 38033 Ross Levine 39 299 25 37600 Christopher Sims 40 24 53 36167 Justin Wolfers 41 21 633 35867 Alberto Alesina 42 209 27 35400 Mark Gertler 43 290 16 35267 Gary S. Becker 44 176 20 35067 Patrick Honohan 45 20 919 34733 Robert J. Barro 46 369 4 33833 Simon Kuznets 47 25 942 33767 David Weinstein 48 42 451 32767 Richard Layard 49 57 544 32167 Michael Kremer 50 166 442 31033 Maurice Obstfeld 51 212 35 30867 Raghuram G. Rajan 52 120 45 30800 Emmanuel Saez 53 28 361 30000

141

Table 7.6 continued

Name Ranking: Ranking: Ranking: Average # Google Bing RePEc of (Average Rank Google Score) pages Thomas J .Sargent 54 67 11 30000 Carmen M. Reinhart 55 180 51 29600 Francesco Giavazzi 56 94 477 29233 David Romer 57 77 121 28700 Paul R. Krugman 58 125 19 27633 Alan B. Krueger 59 118 37 26767 Elhanan Helpman 60 228 24 26033 Jean Paul Fitoussi 61 47 834 25633 Ernst Fehr 62 213 92 25600 James J. Heckman 63 181 2 25100 Michael Greenstone 64 61 968 25067 Richard H. Thaler 65 81 125 25067 T. N. Srinivasan 66 139 452 24867 Xavier Vives 67 132 180 24767 Simeon Djankov 68 152 267 24000 Kenneth J. Arrow 69 59 291 23800 Ashoka Mody 70 391 641 23533 Edward C. Prescott 71 124 17 23167 Sendhil Mullainathan 72 225 406 23100 Guido Tabellini 73 208 108 22867 Alan S. Blinder 74 121 134 22733 Reinhard Selten 75 58 806 22667 Douglass C .North 76 74 420 22167 Martin Ravallion 77 424 66 21833 Ricardo Hausmann 78 92 713 21600 Narayana Kocherlakota 79 13 370 21333 Costas Azariadis 80 458 697 21300 Charles Wyplosz 81 119 582 21300 Eswar Prasad 82 110 468 21100 James H. Stock 83 375 14 21100 Oliver E. Williamson 84 63 692 21067 Enrico Moretti 85 122 479 21033 George A. Akerlof 86 65 67 20933 Richard Blundell 87 304 22 20433 Robert C. Merton 88 62 94 20433 Bruno S. Frey 89 301 81 20067 William D. Nordhaus 90 97 185 20067 Lawrence F. Katz 91 382 49 20000 Robert W. Vishny 92 398 31 20000 Luigi Guiso 93 84 296 19567 Lucian Bebchuk 94 334 851 19467 Sheridan Titman 95 238 238 19467 Campbell R. Harvey 96 447 149 19400 Eduardo Levy Yeyati 97 386 872 19367 Steven Shavell 98 230 491 19200 Zvi Griliches 99 249 88 19167 George Loewenstein 100 93 343 19133

142

Table 6.7 Correlations based on corrected values

Searches Average: Average: Chang Google Bing e

John Bates Clark Medallists “Name” - (“Name” + “John Bates ↓(G) 0.1872 0.2675 Clark”) ↑(B) (0.000) (0.000) “Name” - (“Name” + “John Bates ↓(G) 0.1848 0.2647 Clark Medal”) ↑(B) (0.000) (0.000) ↓(G) “Name” - (“Name” + “JBC Medal”) 0.1779 0.377 ↑(B) (0.000) (0.000) Nobel Prize Winners ↓(G) “Name” - ( “Name”+ “Nobel”) -0.1127 0.3339 ↑(B) (0.0024) (0.000) ↓(G) “Name” - (“Name + “Nobel Prize”) 0.0770 0.3425 ↑(B) (0.0385) (0.000) Notes: G: Google, B: Bing. Using the average values of these two searches, we report the changes from the correlations reported in Table 6.3. The search process for John Bates Clark Medallists was conducted on 27 April and 5 May and for Nobelists on 24 May and 30 May.

143

Chapter 7 Do The Best Scholars Attract the Highest Speaking Fees? An Exploration of Internal and External Influence

Chan Ho Fai, Bruno Frey, Jana Gallus, Markus Schaffner, Benno Torgler & Stephen Whyte

Scientometrics (2014), 101(1), 793-817.

Abstract

This study investigates whether academics can capitalise on their external prominence (measured by the number of pages indexed on Google, TED talk invitations or New York Times bestselling book successes) and internal success within academia (measured by publication and citation performance) in the speakers’ market. The results indicate that the larger the number of web pages indexing a particular scholar, the higher the minimum speaking fee. Invitations to speak at a TED event, or making the New York Times Best Seller list is also positively correlated with speaking fees. Scholars with a stronger internal impact or success also achieve higher speaking fees. However, once external impact is controlled, most metrics used to measure internal impact are no longer statistically significant.

One of the problems of our time is to overcome attitudes that tend to justify and reinforce the isolation of the scientific community. We must open new channels of communication between science and society. Prigogine and Stengers (1984, p. 22)

When we, as scientists, build and use tools and infrastructure that support open dissemination of actionable, accessible and auditable metrics, we will be on our way to a more useful and nimble scholarly communication system. Piwowar (2013, p. 159)

145

‘Publications’ is just one mode of making public and one way of validating scholarly excellence. It is time to embrace the Web’s power to disseminate and filter scholarship more broadly and meaningfully. Welcome to the next era of scholarly communication. Priem (2013, p. 440).

Introduction

A scientist’s primary goal is the advancement of knowledge. Scholarly activity can take on multiple forms, ranging from conducting purely scientific research to offering policy recommendations. Scholars are expected to perform well across several different arenas that can be classified as either internal or external. Research and participation in academic self-governance are internal, while engagement in the general societal discourse (i.e. private institutions, government or civil society) is external to the academic system (see Aguinis et al. 2012). In other words, internal impact refers to the impact of the research upon other scholars (or scholarly organization) within the realm of academia while the latter is the ability to influence any other non- academic institutions or people outside academia. Thus, measures of external impact involve assessment of much broader societal level proxies for influence. Teaching lies in between: its influence is internal as long as the students are at the university, and it is external once the former students are working as professionals in society beyond academia.

However, the impact of scholars has primarily been analysed by investigating internal impact via measures such as citations or publication counts. The use of such metrics dates back to as least as far as the 18th century when publication counts were practiced in the legal field (Shapiro 1992); appeared prominently in a paper on citations by Gross and Gross (1927) in Science; experienced a sharp surge in the late 1960s (Glänzel and Schoepflin1994); and still are becoming increasing popular in evaluating the performance of scientists (Radicchi et al. 2008). Inherent in the counting of publications and citations we find the most fundamental social processes of science: success in communicating and exchanging research findings and results (Fox 1983). It is seen by many as the ultimate indicator of effectiveness (Certo et al. 2010). However, Henrekson and Waldenström (2011), for example, asked the question: “Should we give weight to research’s impact outside academia, such as influence on policy-making or the policy

146 debate? (p. 1154)”. Aguinis et al. (2012) paraphrase Donald C. Hambrick, a former Academy of Management president, when criticising that “the way we currently assess the impact of our scholarly work seems to be based on an incestuous, closed loop” (p. 106). According to Aguinis et al. (2012), even the applied area of management has achieved limited success in making a substantial impact on stakeholders outside the university. The limited research available reveals no, or only a low, correlation between external and internal influence among scholars in management (Aguinis et al. 2012) and economics (Chan et al. 2013). Those studies look at the correlation between academic performance (publications, citations) and external influence as measured by the number of pages indexed on the search engine Google. Quantifying performance based on pages indexed on the Internet is incomplete as what it does not take into account is the relative value of the information on such pages. One single page referencing a scholar’s name may be of great importance, but it may also be close to meaningless. We therefore extend the previous research by approximating the relevance of a scientist’s contributions by measuring an audience’s willingness to pay in the speakers’ market.

This paper analyses whether scholars better known to the general public earn higher speaking fees than scholars with superior research performance within academia. We find that both external prominence and internal impact can be capitalised on the speakers’ market although external prominence has a stronger impact. Once the number of pages indexed by Google is controlled (excluding Google.edu entries) as an external influence proxy, most internal impact factors lose their statistical significance.

Provided the willingness to pay for scholars in the speakers’ market correctly reflects the relevance of their knowledge for practical issues, our results raise the question of the extent to which academic research is of interest to the public (see, e.g., Frey 2006; Van Bergeijk et al. 1997; Van de Ven and Johnson 2006). The findings may be seen as an indication of a considerable gap between scholarly research and practical pertinence. This paper is organised as follows: “Scholarly impact” section aims at clarifying the motivation of the current analysis by providing a short overview of what we know so far regarding scholarly impact. Section “Data” describes the data collection process. Estimation results are presented in “Estimation results”

147

section followed by the discussion and conclusion in “Discussion and conclusions” section.

Scholarly impact

Several countries such as the United Kingdom, Australia, Italy, New Zealand, and Germany have moved towards a national performance-based research evaluation model, which employs impact factor as an instrument for assessment (Cameron 2005; Thelwall 2008; Owens 2013). Such metrics are also used in hiring and tenure decisions. Committee members often struggle with the task of reading an entire body of work when academics are considered for promotion. This increases the incentive to take shortcuts (Priem and Hemminger 2010) and assess a candidate based only on the journal impact factors (Cameron 2005). Moreover, due to article overload (Priem and Hemminger 2010; Torgler and Piatti 2013), readers searching for articles will also look for shortcuts, and therefore employ identifiable indicators such as journal quality or former article citation performance to evaluate whether it is worth looking at an article. However, the US National Science Foundation has started to ask principal investigators to list their “products” rather than “publications” in the biographical overview, which takes into account the breadth of intellectual possibilities such as data sets, software, patents, copyrights or other non-traditional products (Piwowar 2013).

Van Raan (1997) states that one of the core interests of scientometric research is the “development of methods and techniques for the design, construction, and application of quantitative indicators on important aspects of science and technology” (p. 206). Thelwall (2008) points out that in the last 50 years there have been two major changes in the way research can be quantitatively analysed. The first was the creation of the Institute for Scientific Information database in 1962 and the second was the Web publishing covering a broad range of research-related documents. We can now observe a renaissance of new initiatives attempting to measure scientific impact, from usage log data, to distributional statistics, and sophisticated social network approaches (Bollen et al. 2009). Thelwall and Harries (2004) are critical of how quantitative research has historically been restricted to formal communication (journals and books), but the Web allows new

148 complementary approaches to be derived. Many altmetrics have emerged for tracking public attention. For example, journals such as PLoS ONE have started to provide metrics for each online published article (views, downloads, cites, saves and discussions (e.g., Twitter, Facebook, Comments, Google blogs). Receiving attention is a sign of success (Franck 1999), and the Web is able to collect previously hidden information on these signs of success, while allowing scientific information to be distributed widely via videos, slides, blog posts, Twitter etc. (Letierce et al. 2010). Data repositories such as figshare track downloads are increasing in popularity. In addition, there is a tendency to extend the traditional measures of scholarly importance beyond the academic environment. Piwowar (2013) argues that these kind of “altmetrics give a fuller picture of how research products have influenced conversation, thought and behaviour. Tracking them is likely to motivate more people to release alternative products—scientists say that the most important condition for sharing their data is ensuring that they receive proper credit for it” (p. 159). The editorial of the renowned scientific journal Nature (2013) has emphasized: “The conventional measures of scholarly importance—citation metrics, publication in influential journals and the opinion of peers as expressed in letters and interviews—still loom large. But to those are now added metrics such as article downloads and views, and measures of importance beyond the academic realm, including influence on policy-makers or health and environmental officials, effects on industry and the economy, and public outreach” (p. 271). Holbrook et al. (2013) published a list of 33 positive and negative indicators of impact in Nature, differentiating between public engagement, academic community, and the media.

Reward mechanism Scientists play an important role in the creation and dissemination of knowledge in society (Gomez-Mejia and Balkin 1992). However, it is not clear whether scientists are actually paid according to the importance of this role. So far, the main focus of the literature has been on the determinants of university pay. However, fiscal constraints at universities produce interest in exploring opportunities for financial reward outside the academic community. The determinants of a society’s reward structure has been identified as an important question (Acemoglu 1995).

The fee a speaker can command is a professional and commercial assessment of academics’ or public figures’ worth in the marketplace. It is a

149 method of quantifying how the ability of these figures is judged beyond academic success, and how society evaluates the (professional) importance and interest of speakers (societal appraisal). We currently have no understanding as to what drives success in this market. The literature has primarily maintained an intense focus on how internal impact influences academic salaries, finding a positive correlation between academic performance and salary (see, e.g., Katz 1973; Hansen et al. 1978; Hamermesh et al. 1982; Diamond 1986; Kenny and Studley 1995; Moore et al. 1998; Bratsberg et al. 2003; Duncan et al. 2004). Performance proxies are linked to reward systems in academia as actual effort or intentions cannot be observed publicly (Dasgupta and David 1994). Diamond (1986), for example, asked the question: “What is a citation worth?” After summarising several studies, he concludes that citations have positive and significant effect on earnings. For example, he demonstrates that the marginal value of a citation when the level of citations is zero is between $50 and $1,300. However, the marginal value depends on practices within disciplines. For example, the quantity of publications and citations tend to be relatively low in disciplines such as economics and mathematics, hence a citation’s marginal value is higher when compared to fields such as chemistry or physics. Kenny and Studley (1995) found that publications and citations together account for 20 percent of the salaries in their data sample. Public speaking engagements by academics give business conferences and meetings both internal and external status and credibility. It comes as no surprise that some academics now advertise their speaking services globally across a range of public speaking and toastmaster websites. In this competitive market, speaking fees reveal the willingness to pay according to the perceived value or contribution of a scientist, and the public interest in the commodity that the scientist provides. This paper seeks to quantify what influences the attractiveness (and hence the commercial value) of scientists.

Social effects An understanding of what influences speaking fees requires a closer look at the importance of external impact. In recent times there has been a stronger emphasis on accounting for the social effects of science. For example, Frodeman and Holbrook (2007) report that in 2001 the National Science Foundation informed scientists that failing to address the connection between research and its broader effects on society in grant proposals would result in

150 the proposal being returned without review. Since the 1980s, the UK Economic and Social Research Council requires its grant applicants to demonstrate how they will deliver practitioner-relevant outputs (John 2012).

Dasgupta and David (1994) states: “To say what goes on within the sphere of human activities identified as ‘The Republic of Science’ has grown too important for the rest of society to leave alone is also something of a commonplace assertion” (p. 488). Frodeman and Holbrook (2007) stress that “it is no longer accepted that scientific progress automatically leads to societal progress” (p. 29). John (2012) notes that: “There is a revolution in information afoot whereby anyone can produce output that can feed swiftly into public debate. The rapid development of the internet, in particular social media such as Twitter, weakens the power of traditional gatekeepers and creates opportunities for entrepreneurial advocates and communicators” (p. 18). The social function of science is not a new topic (see, e.g., Bernal 1939). In discussing Bernal’s book, Merton (1941) points out: “More recently, a changing social structure, which aroused Frankensteinian guilt-feelings and a correlated sense of social responsibility, has induced a considerable body of scientists to consider the social role of science” (p. 622).

Scientists have been criticised for misunderstanding media (Crichton 1999). Crichton, a renowned writer and film director suggests: “You need working scientists with major reputations and major accomplishments to appear regularly on the media, and thus act as human examples, demonstrating by their presence what a scientist is, how a scientist thinks and acts, and explaining what science is about” (p. 1463). Many scientists have expressed the concern that popularization would reduce their status among their peers (Willems 2003). The environment does not provide sufficient encouragement to be involved in public dissemination of information (Dunwoody and Ryan 1985). However, survey results obtained by Peters et al. (2008) discover that researchers see increasing the public’s appreciation of science as the most important reason to motivate interaction with the media. Kidd (1988, p. 127) argues that one of the most compelling arguments in favour of popularization of science is an increasing proportion of public policy decisions that have a scientific or technical aspect while the wellbeing of society depends on well-informed citizenry.

151

There is a trend among researchers and academic institutions towards establishing a stronger tie between science and society (Jensen et al. 2008). The Royal Society dedicated in 2006 a report entitled Science Communication to survey factors affecting science communication by scientists and engineers. Martin Rees (at that time the President of the Royal Society) emphasized in his foreword that “Scientists need to engage more fully with the public. The Royal Society recognises this, and is keen to ensure that such engagement is helpful and effective. The role of science in public policy is becoming even more pervasive. Many scientists are willing to engage in dialogue and debate, but they need encouragement and guidance, 35 and they need to feel that their efforts are valued”. In 2013, PNAS, the flagship journal of the National Academy of Sciences organised a special issue on the science of science communication edited by Baruch Fischhoff and Dietram Scheufele. They state that “[m]aking the most of what science has to offer society requires the give-and-take of two-way communication with laypeople. Those interactions can be direct, as in classrooms and social settings, or indirect, through the mediation of research helping scientists to understand the public and vice versa” (Fischhoff and Scheufele 2013, p. 14031). The LSE Public Policy Group has even developed a handbook for 36 social scientists entitled “Maximizing the Impacts of your Research”. The second part of the report is entirely dedicated to generating impact beyond academia. A study using records of more than 3500 scientists over a three- year period (2004–2006) in France indicate that dissemination activities are neither bad nor good for scientists’ careers. However, it has a positive effect on promotions. In addition, scientists who are more engaged in dissemination are also more academically active (Jensen et al. 2008).

Data

Speaking fees It is very difficult to consistently measure speaking fees paid to scholars. Systematic data on the remuneration for such activity across countries is limited (Hosp and Schweinsberg 2006). Many, if not most, academics do not

35 http://www.ulb.ac.be/inforsciences2/communication/coursComm/docs/royal_soci ety.pdf. 36 http://blogs.lse.ac.uk/impactofsocialsciences/book/.

152 ask for any money if they are invited to present a keynote address to a scientific society or to give a lecture at a research seminar. In contrast, they often try to maximize their remuneration if they are invited by a for-profit institution.

The data regarding speaking fees were collected during December 2013 and January 2014 from the website of eight speaking agencies, including BigSpeak Speakers Bureau,37 Keppler Speakers,38 Leading Authorities Speakers Bureau,39 Premiere Motivational Speakers Bureau,40 41 42 43 Speakerpedia , Speakers Platform, the Sweeney Agency and Washington 44 Speakers Bureau. Each site lists the speaking fees, the biography and the contact information of the speaker. In most cases, the speaking charges are listed in a range (e.g., $7,500–$12,500 or $50,000?), since the exact charges might vary due to the location and type of invitation. Hence, we take the lower end of the fee range to consistently measure the minimum speaking charges 45 for the speaker (unless no range is specified). In addition, if a speaker is listed on more than one agency website, we use the highest of the lower end range values as the speakers’ charge. The minimum speaking fees reported range from $750 to $250,000. We exclude from the sample those speakers who do not disclose speaking fee charges. Moreover, in addition to the biography of the speakers, the speaking agencies also classify speakers into different categories, for example, Speakerpedia classifies speakers into twelve categories such as “Arts and Humanities”, “Business”, “Government & Policy”, and “Internet & Technology”. With such information, we are able to differentiate academic speakers from non-academic speakers working in specific fields by using a set of strings as filter.46

37 http://bigspeak.com/. 38 http://kepplerspeakers.com/ 39 http://leadingauthorities.com/ 40 http://premierespeakers.com/ 41 http://speakerpedia.com/ 42 http://speakersplatform.com/ 43 http://thesweeneyagency.com/ 44 http://washingtonspeakers.com/ 45 Only 3.79 % (22 speakers) have no fee range. 46 We first use the combination of the words “professor”, “director”, “fellow” and suffix such as “-mist”, “-logist”, “-icist” with words like “university”, “college”, or “institute” to identify whether the speaker is a scientist. Then, we search for prefix such as “econ”, “bio”, “phy”, “psy” and “medic” to classify speakers into in the fields in which they are active. To ensure the accuracy of this automated filtering, we perform a manual Google search on the career field of speakers who are filtered out.

153

Furthermore, with automated word searches within speakers’ biographies, we are able to classify speakers’ fields into three disciplines, namely, (1) natural science (e.g., biology, medicine and physics), (2) social science (business) (e.g., economics, finance, management and marketing) and (3) social science (others) (e.g. psychology, sociology, and politics). It is important to differentiate between fields as disciplinary practices and conventions affect the scientist’s role in the production and dissemination of knowledge, and influences how information and communication technologies and the Internet are used (Barjak et al. 2007). Next, we categorise speakers into three categories according to the degree of academic involvement. By 47 examining the speakers’ CVs, LinkedIn and Wikipedia pages, we define someone as a “full-time” academic if the speaker has spent more than half of his/her career as a researcher in an academic institution or organization. On the other hand, a “part-time” academic has spent more than half of his or her career in a private or government institution while still being affiliated for some of the period in academia. We then define a speaker as non-academic if he or she has never worked in or been affiliated with an academic institution. Table 7.1 provides the sample sizes of these different categories and Figure 7.1 depicts the distribution of minimum speaking fees by the degree of academic involvement and professional field. We observe that all distributions of speaking fees are positively skewed. The speaking fee distribution between academic and non-academic speakers is not significantly different although mean value is higher for academic speakers (p value of the Kolmogorov–Smirnov test equals 0.076). Among academics, the distributional difference between part-time and full-time academics is also not significant (p = 0.311). However, we find that the distribution of speaking fees for academic speakers who are in business related disciplines is different from academics with natural science (p\0.000) and other social science background (p\0.000), and the latter two disciplines share similar distribution (p = 0.877). A similar pattern is observable between disciplines when focusing only on academics.

47 https://www.linkedin.com/

154

Table 7.1 Sample size by academic involvement and fields

Academic Natural Social Science Social science Total Involvement /Fields Science (Business) (Others) Part-time academic 31 61 32 124 Full-time academic 52 88 50 190 Non-academic 73 127 66 266 Total 156 276 148 580

Figure 7.1 Distribution of minimum speaking fees

155

External influence Number of pages as indexed by Google

The Web is becoming crucial as an information interface, hosting a large variety of academic information (Thelwall and Price 2003). The external importance of a speaker (prominence) in the public is measured by the number of web pages referring to a speaker’s name. It is also an indicator of visibility. We first conducted an automated search on 14 April, 2014 via the Google search API (application programming interface) to obtain the 48 number of hits. In addition, we obtained the number of non-education domain web pages to reflect the external prominence of academics. To capture the most accurate number of searches, we used the publication names for speakers who have published a book or a scientific article, typically appearing in the form of “[first name] [initial of the middle name] [surname]” or just “[first name] [surname]”. A double quote is placed before and after the search item, i.e. the speaker’s name, to generate results of the exact search phrase. The first 50 pages returned were then manually checked to identify names with spurious matches. If five or more pages (10 %) were not attributed to the speaker, we excluded the person from the sample, which resulted in a 49 total sample size of 580 speakers. Google index pages afford the opportunity 50 to also cover some of the more informal scholarly communication (all other forms of communication beyond publications).

The left-hand side of Figure 7.2 shows the distribution of the number of indexed pages. As evidenced, the distribution is highly skewed. The right- hand side shows what we would call a “Google index page Lorenz curve,” thereby providing an impact inequality proxy for all the speakers. This figure reveals a significant level of impact inequality (Gini coeff. = 0.85); for example, 20 % of the speakers are responsible for almost 90 %of the indexed pages.

48 There are, of course, other possible methods by which we could measure external impact. For an overview, see Chan et al. (2013). 49 154 speakers are excluded from all 734 eligible speakers filtered out by the automated search due to spurious name matches. 50 For a discussion regarding informal and formal communication see Kousha and Thelwall (2007).

156

Figure 7.2 Impact inequality

TED talks

It is something of an understatement to say that the Internet has become a very important source of information. The Science and Engineering Indicators 2014 provided by the National Science Foundation reports that around 4 in 10 Americans cited the Internet as their primary source of science and technology information in 2012.51

We are going to take a closer look at TED talks who have become internationally very famous. TED started in 1984 as a conference for speakers in the area of technology (T), entertainment (E), and design (D) to discuss their best ideas (Rubenstein 2012). TED’s mission is to build “a clearinghouse of free knowledge from the world’s most inspired thinkers” (http://www.ted.com/pages/about/). The most popular TED talk as of 52 December 2013 was Sir Ken Robinson’s How Schools Kill Creativity (Feb 53 2006) with 23,510,221 views. In 2012, TED reached 1.5 million views per day.54

For our analysis, we check whether the speakers were invited to present at conferences held by TED Conferences, LLC before 2013. This includes the main TED conferences, TEDGlobal, TEDMed and other TEDx

51 http://www.nsf.gov/statistics/seind14/index.cfm/chapter-7/c7h.htm 52 http://www.ted.com/talks/ken_robinson_says_schools_kill_creativity 53 http://blog.ted.com/2013/12/16/the-most-popular-20-ted-talks-2013/. As of April 30, 2014 it had attracted 26,222,340 views. 54 http://blog.ted.com/2012/11/13/ted-reaches-its-billionth-video-view/

157

55 events. Surprisingly, to the best of our knowledge, the implication of TED talks has hardly been analysed so far in academic journals (exceptions are Sugimoto and Thelwall 2013; Rubenstein 2012). Sugimoto and Thelwall (2013) point out that online videos provide a novel platform for popularising science, and that tracking online interaction can act as attention metrics that could feed into new forms of academic capital. Moreover, positive affirmations in this environment may encourage scholars to be active on these platforms.

Books and their recognition

In a study of 148 full professors in economics, Hamermesh et al. (1982) were not able to detect a positive impact of book publication on earnings. On the contrary, in one specification, the coefficient of the variable ‘books’ was even negative and statistically significant. However, Katz (1973) studied a substantial cross-section of disciplines and found that books had a positive influence on salary. The publication of a book was worth an extra $230 in a professor’s lifetime while an extra publication was worth $18. Finkenstaedt (1990) discusses the measurement of research performance in the humanities, highlighting the importance of books: “And the book addressed to the good old common reader is probably more valuable for society than a specialised article—in spite of its many citations. It may even be that case that the common reader does not get the books he deserves because a mistaken idea of “impact” makes the junior staff publish articles instead of readable books” (p. 414). Bratsberg et al. (2003) employ a large dataset of 1897 observations on 176 faculty members at five universities and observe a positive, highly statistically significant relationship between books and earnings. On the other hand, Melguizo and Strober (2007) also analyse a large data set, and did not find that books have a significantly positive impact on faculty salary determination for any of the fields studied (Science, Engineering, Professional, Social Sciences, Education, Humanities and Arts). Yet, articles in refereed journals did return a significant positive impact (with the exception of Humanities and Arts).

55 The list of TED speakers is extracted from https://www.ted.com/talks/list/.

158

First, we count by speaker the number of books that were listed on 56 the Library of Congress of the United States, including books that are published or translated into languages other than English. Books are an alternative to journal articles and are able to target an audience beyond academia. The number of titles available on the online catalogue of the Library of Congress of the United States has been used in the past as a measure of long-term quality (Ginsburgh 2003).

In addition, to obtain a better proxy for success we measure the external influence of books by using the New York Times Best Seller list and various non-fiction book prizes. We obtain two proxies from Hawes 57 Publications, which has documented the top 15 bestselling books of the weekly NYT bestseller list in both fiction and non-fiction since 1st January 1950. Our proxies are the number of NYT best-selling books written by the speaker, and the number of weeks these books have stayed on the list.

Moreover, we collect book award data for several major non-fiction 58 book awards, including the Pulitzer Prize and the National Book Award. The following book awards were obtained by at least one speaker from our sample (number of winners in bracket): Lannan Literary Awards (1), Los Angeles Times Book Prize (4), Michael Faraday Prize (2), National Book Critics Circle Award (1), Pulitzer Prize (4), Royal Society Prizes for Science Books (2) and Science Communication Awards (1).

Descriptive analysis We first take a look at the differences between academics and non-academics (Table 7.2). While the speaking fees are higher for academics, the external impact is larger for nonacademics. However, the difference in external impact is only statistically significant if we exclude the.edu domains from the counts. Academics produce more books and are more frequently engaged in TED

56 See http://catalog.loc.gov/. 57 See http://www.hawes.com/pastlist.htm/. Prior to 11th September 1977, the best- selling list captures the top 10 best-selling books. 58 The list includes (in alphabetical order): Anisfield-Wolf Book Awards, Boston Globe–Horn Book Award, Dingle Prize, Donald Murray Prize, Financial Times and Goldman Sachs Business Book of the Year Award, Heartland Prize, Innis-Gérin Medal, Jerusalem Prize, Kistler Prize, Lannan Literary Awards, Los Angeles Times Book Prize, Ludwik Fleck Prize, Michael Faraday Prize, National Book Award, National Book Critics Circle Award, Pulitzer Prize, Royal Society Prizes for Science Books, Samuel Johnson Prize, Science Communication Awards, Science in Society Journalism Awards, and the Specsavers National Book Awards.

159 talks (all the metrics are statistically significant). They are also acknowledged with awards more frequently while the average number of NYT bestsellers is almost identical. However, books by academics tend to remain on the list for longer.

Table 7.2 Academics versus non-academics

Proxies Non-academic speakers Non-academic speakers t-test

Mean SD Min Max Mean SD Min Max

Male 0.72 0.45 0 1 0.77 0.42 0 1 -1.54

Professional Age 33.04 13.24 3 71 31.48 12.33 0 64 1.45 log(Google page) 8.07 2.31 0 13.86 7.77 1.97 2.08 13.53 1.64 log(Google page) 8.06 2.30 0 13.85 7.65 2.05 1.95 13.52 2.26* exclude .edu domain Minimum Speaking 18309 19788 1000 250000 20617 21150 750 200000 -1.79 Fee Number of books on 6.97 15.22 0 174 9.43 17.11 0 213 -2.30* Library of Congress

TED talk speaker 0.07 0.26 0 1 0.17 0.37 0 1 -3.67***

Number of times 0.08 0.31 0 2 0.25 0.73 0 9 -3.46*** invited to TED Non-fiction book 0.004 0.06 0 1 0.03 0.18 0 1 -2.48* award dummy NYT Best Sellers 0.18 1.16 0 18 0.17 0.62 0 5 -0.05 (number of books) NYT Best Sellers 1.02 7.05 0 107 1.67 9.46 0 126 -1.17 (number of weeks) Note: The symbols *, **, *** represent statistical significance at the 5%, 1% and 0.1% levels, respectively.

In Table 7.3 we present the correlation between these different external influence proxies. As evidenced by the results, the correlation is not very high, indicating that they are measuring different aspects of external influence which justifies the collection and exploration of these different proxies.

160

Table 7.3 Correlation matrix of external influence proxies

log(Google # of books TED TED talk Book NYTBS NYTBS Correlation log(Google p.) p. w/o .edu) on LoC talk (times) prize (# books) (# weeks)

log(Google p.) 1

log(Google 0.999*** 1 p. w/o .edu)

# of books 0.333*** 0.329*** 1 on LoC

TED talk 0.172*** 0.169*** -0.024 1 TED talk 0.180*** 0.178*** -0.032 0.770*** 1 (times) Book prize 0.104* 0.104* 0.234*** 0.061 0.106* 1

NYTBS 0.218*** 0.217*** 0.425*** 0.058 0.056 0.081* 1 (# books)

NYTBS 0.170*** 0.169*** 0.310*** 0.123** 0.127** 0.125** 0.746*** 1 (# weeks) Note: Correlations over 0.4 are in bold. The symbols *, **, *** represent statistical significance at the 5%, 1% and 0.1% levels.

Internal influence Internal indicators such as citations can be seen as way of measuring collegial reputation (Reskin 1977) or whether contributions are broadly relevant to the scientific enterprise (Merton 1973) and in particular to the current research frontier (Diamond 1986). They are increasingly available not only within the academic profession but also to the general community through avenues such as Google Scholar. To derive the internal influence metrics we rely on Publish 59 or Perish 4, a software program that retrieves raw citations using Google Scholar or Microsoft Academic Search and analyses those citations using a broad set of measures. As Google Scholar offers better coverage (also for a cross-disciplinary comparison), we will only work with Google Scholar data. Another advantage of Google Scholar is that it covers the newer material better (Bauer and Bakkalbasi 2005) as well as material beyond peer-reviewed journal contributions (Levine-Clark et al. 2008) that could be relevant in measuring the internal impact of scholars (e.g., working papers). Bar-Ilan (2010) concludes that “Google Scholar’s coverage was surprisingly good, and its accuracy was also better than expected” (p. 506).

59 See http://www.harzing.com/pop.htm/ and Harzing (2010).

161

We will focus on the following metrics: number of papers, total citations, average number of citations per paper, h-index, and hIa-index. The total number of papers is a quantitative measure that takes into account productivity but not the importance or impact of the papers (Hirsch 2005). On the other hand, the h-index provides a measure described by Hirsch (2005) as “an estimate of the importance, significance, and broad impact of a scientist’s cumulative research contributions” (p. 16572). It therefore incorporates both quantity and visibility of contributions (Bornmann and Daniel 2007). However, the h-index has been criticised for being insensitive to outstandingly highly cited papers (Egghe 2006) and punishes newcomers who 60 have both a low publication output and number of citations (Glänzel 2006). Thus, we also explore the average number of citations per paper, allowing a comparison between scientists of different ages (Hirsch 2005). Harzing et al. (2014) are also critical of the problems with the h-index when comparing academics working in different disciplines, due to dissimilar publication and citation backgrounds. They develop a new metric, termed hIa-index that allows for a more reliable comparison between academics in different disciplines and at different career stages. The authors correct for a considerable part of the variation across disciplines by using the number of coauthors61 and to correct for differences in career length they take into account the numbers of years an academic has been publishing. To be more precise, they first normalise citations for each paper by dividing the number of citations by the number of authors for that paper. Next, they calculate the h-index which is now based on normalised citation counts. Finally, they divide this value by the number of years that an academic has been publishing. Harzing et al. (2014, p. 818) recommend to use this index in conjunction with the h-index and the total number of citations. We will also report the total number of citations which has the advantage of measuring the total impact (Hirsch 2005). However, such a measure could be driven by a small number of outstanding contributions (Hirsch 2005).

Referring to also other studies, Bar-Ilan (2010) states: “currently, there is no single citation database that can replace all the others” (p. 505). Thus, as a robustness test, we record the total number of publications and

60 See Glänzel (2006) for a further discussion of shortcomings of the h-index. 61 Publish or Perish limits the maximum number of authors considered to 50 (Harzing et al. 2014).

162 citations, and the average citations received (without self-citation) from Scopus. The correlation matrix of the variables reported in Table 7.6 indicate that, the correlation between the different citation metrics is in general not that high which shows that we are measuring different aspects of internal impact.

We decided to use Scopus instead of Web of Science as Scopus uses a valuable author identifier. The database employs an algorithm that matches authors based on several characteristics such as affiliation, address, subject area source title, dates of publication citations, and co-authors (Li et al. 2010). With more than 23,67462 journals (around 6600 more journals that the Web of Science63 databases), Scopus offers a greater breadth of coverage than Web of Science (Levine-Clark et al. 2008). In addition, Scopus has a very user friendly interface (Li et al. 2010).

Employing a set of different indicators takes into account the fact that scientists have different career paths and comparative advantages. Dixit (1994, p. 12) has nicely pointed this out: “Some people are good sprinters in research. They can very quickly spot and make a neat point; they do this frequently, and in many different areas and issues… In the same metaphor, others are middle-distance runners… A few… are marathoners; they run only a small number of races, but those are epics, and they get the most (and fully deserved) awe and respect. In contrast, the profession seems to undervalue sprinters. But each kind of work has its own value, and the different types are complements in the overall scheme of things. Progress of the subject as a whole is a relay race, where different stretches are of different lengths and are optimally run by different people”. Hirsch (2005, p. 16571) also points out that “a single number can never give more than a rough approximation to an individual’s multifaceted profiles, and many other factors should be considered in combination in evaluating an individual”.

62 See http://www.elsevier.com/online-tools/scopus/content-overview/. 63 http://wokinfo.com/citationconnection/ and http://ip-science.thomsonreuters .com/cgi-bin/jrnlst/jlresults.cgi? PC=MASTER.

163

Estimation results

The relationship between external prominence and minimum speakers’ fees is plotted in Figure 7.3. The nonlinear structure of web index entries is taken into account by showing the results in log(indexed web pages) by Google on the left-hand side and log(indexed web pages excluding.edu) on the right- hand side. The figure demonstrates a positive relationship with a correlation of 0.365 (Pearson’s r) left and 0.363 right suggesting that external prominence may impact the ability to obtain high rents on the market for speaking fees. Figure 7.4 in the Appendix depicts the results with sub-samples. Speakers with a part-time or no academic involvement have a higher correlations (r = 0.414 and r = 0.422 for non-academic and part-time academics, respectively) compared with full-time academics (r = 0.309). Therefore, it seems that at least in monetary terms, non-academic speakers are better able to capitalise on their external prominence.

Figure 7.3 External prominence and minimum speaking fee

Table 7.4 reports the results of eight OLS regressions. In each specification, we control for academic involvement, gender, professional age, and field. Professional age is approximated by the career length of the speakers (year since their highest education). In addition, we create a dummy variable to identify speakers who have earned a doctorate degree (e.g. Ph.D., DBA, J.D. or M.D. etc.). We also control for gender differences. Gender differences in salary differentials in the academic labour market have been an important topic in the literature. For example, Barbezat (1987, 1991) observes

164 that salary discrimination has fallen since 1968. In addition, we have also controlled for the location of the speaker (42 speakers are located outside North America: 8 in Asia and 34 in Europe) and whether the speaker has won the Nobel Prize. The Nobel Laureates in our data set are Jimmy Carter (Peace); Steven Chu (Physics); and Robert Mundell, Myron S. Scholes, Daniel Kahneman, Amartya Sen and Robert W. Fogel (all in Economics).

With respect to our key variables, the indexed web pages report a strong influence on speaking fees. The coefficient is highly statistically significant and speaking fee elasticity suggests that a 1 % increase in indexed web pages increases minimum speaking fees by 0.161 % in specification (1) and 0.159 % in specification (2). In specification (2) we subtracted the number of web pages in education domains (URL which contains the domain.edu) from the total Google page count to provide a more accurate measure of external impact, since pages in education domains refer to internal academic activities. Next, we analyse the effect of books. The coefficient of the number of books listed in the Library of Congress is also positive and significant at the 1 % level. Writing a book increases the speaking fees by 0.06 %, so therefore the effect is not very large. However, if a book became a NYT best seller, the effect is substantially larger (15.7 %). In addition, the number of weeks on the NYT list also contributed to the market value of a speaker. The effect, ceteris paribus, of each additional week is 1.2 %. TED appearances are also correlated with higher speaking fees. The dummy variable for having given a TED talk and the number of invitations are highly statistically significant. Those who have spoken at a TED gathering have on average (when holding other factors fixed) substantially higher speaking fees (26 percent). On the other hand, authoring a non-fiction award-winning book does not increase the minimum speaking fee at a statistically significant level.

Looking at the control variables we find the tendency for academic speakers ceteris paribus to charge a higher speaking fee relative to their non- academic counterparts with similar external influence. Part-time academics reap the strongest benefit, reporting a coefficient that is always statistically significant. All else being equal, they generate between 20.4 and 32.8 percent higher speaking fees than non-academics. We also observe differences across fields. Social science speakers in the business area have a significantly higher market value than natural science speakers (around 30 %) while the difference between other social sciences and natural science is not statistically

165 significant. Interestingly, there are no gender differences, however seniority matters. An increase in professional experience by 10 years increases the minimum speaking fees by around 10 %. There is also a positive correlation between being a Nobel laureate and speaking fees, although the coefficient is not statistically significant. Interestingly, North American speakers report lower speaking fees. However, when we control for Google impact or TED appearance, the coefficient loses its statistical significance.

166

Table 7.4 Speaking fee and external influence

Dep. Var.: Log(Min. Speaking Fee)

(1) (2) (3) (4) (5) (6) (7) (8) Academics Part-time 0.326*** 0.339*** 0.210* 0.216* 0.204* 0.214* 0.213* 0.214* (4.06) (4.21) (2.42) (2.48) (2.35) (2.46) (2.47) (2.47) Full-time 0.135 0.156* 0.102 0.098 0.089 0.117 0.138* 0.118 (1.91) (2.20) (1.30) (1.25) (1.14) (1.50) (1.80) (1.53) Background

Male 0.007 0.012 0.063 0.061 0.049 0.066 0.069 0.071 (0.09) (0.16) (0.80) (0.77) (0.62) (0.83) (0.88) (0.90) Professional age 0.006** 0.007** 0.008** 0.010*** 0.010*** 0.009*** 0.009*** 0.009*** (2.89) (2.96) (3.02) (4.10) (4.15) (3.81) (3.63) (3.72) Nobel Prize Laureate 0.363 0.357 0.358 0.529 0.537 0.505 0.089 0.315 (1.18) (1.16) (1.04) (1.58) (1.61) (1.50) (0.25) (0.92) North American based -0.106 -0.100 -0.249 -0.206 -0.149 -0.247 -0.263* -0.255* (-0.89) (-0.84) (-1.93) (-1.58) (-1.13) (-1.91) (-2.06) (-1.99) Fields Social science (Business) 0.278*** 0.278*** 0.272*** 0.311*** 0.330*** 0.284*** 0.291*** 0.281*** (3.77) (3.77) (3.38) (3.76) (4.04) (3.50) (3.65) (3.50) Social science (Others) -0.082 -0.087 -0.002 0.042 0.050 0.026 -0.011 0.003 (-0.96) (-1.02) (-0.02) (0.45) (0.54) (0.28) (-0.12) (0.03)

167

Table 8.4 continued

Dep. Var.: Log(Min. Speaking Fee)

(1) (2) (3) (4) (5) (6) (7) (8) Proxies log(Google page) 0.158*** (10.77) log(Google page) excluding .edu 0.156*** (10.72) Number of books on Library of Congress 0.005* (2.37) TED talk speaker 0.227* (2.11) Number of times invited to TED 0.202*** (3.46) Non-fiction book award dummy 0.320 (1.28) NYT Best Sellers (number of books) 0.155*** (4.05) NYT Best Sellers (number of weeks) 0.012** (3.01) N 580 580 580 580 580 580 580 580 Prob. > F 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 R-squared 0.237 0.236 0.091 0.089 0.101 0.085 0.108 0.097 Notes: The symbols *, **, *** represent statistical significance at the 5%, 1% and 0.1% levels, respectively; t statistics in parentheses. The reference group for full-time and part-time academics is non-academics. The reference group for social science (business) and social science (others) is natural science.

168

Moreover, by repeating the analysis in Table 7.4 but restricting the sample to study only non-academic speakers, we find that the effect of log(Google page), number of books and NYT Best Sellers remain significantly positive, but book prize and TED talk (both dummy and number of invitations) are not significant. The positive effects of career age and business discipline also remain statistically significant at 1 % (results not shown). In addition, we investigated whether a PhD is correlated with higher speaking fees and discovered this was not the case. In fact, we actually observe the opposite effect. There is a negative relationship between having a PhD and size of the speaking fees.

In addition, we explore the interaction term between dummies for academic engagement and external impact (log(Google page) with and without exclusion of.edu). The interaction effects are not statistically significant indicating that academics are not more able to profit from their external impact than non-academics.

Next, we analyse the academic performance as captured by Publish or Perish (version 4) and Scopus data on publications and citations. Thus, we restrict our sample to include only those speakers in academia (N = 314). We perform a single regression on each performance measure, controlling for academic involvement, gender, career age, academic field, Nobel Prize and location. We construct two sets of regressions, with and without controls for log(non-education Google pages). Hence, Table 7.5 summarises the results of two sets of 14 OLS regressions. Most of the publication and citation metrics are able to explain the variation in speaking fees. All coefficients are significantly positive for Google Scholar, indicating that scholars with higher internal impact could capitalise their internal success via speaking fees. On the other hand, when using Scopus, only the average citation count is statistically significant. Thus, there is a trend in the results indicating that internal impact matters. However, when we control for log(non-education Google pages) as a key proxy for external influence, the internal impact largely disappears. It remains only for the average number of citations per paper. Furthermore, when analysing the data from Scopus, the coefficient for citations per publication is no longer statistically significant (t = 1.57). One reason could be the accessibility of the data. Google Scholar has a definite

169 advantage in that it is not a commercial-based citation source such as Scopus and is therefore available to anyone. This could be an important point when looking at speaking fees. In addition, the results obtained for citations per publication could indicate that quality matters more than quantity, or a mix of quantity and quality. On the other hand, Google pages are always highly statistically significant. Next, we calculate the standardised/beta coefficients to explore the relative strength of external and internal success. The results indicate that the effect of Google pages is twice (Google Scholar) as strong as for citations per paper. This demonstrates that external impact is substantially more important than internal impact for speaking fees.

Table 7.5 also reports the influence of other external impact proxies on speaking fees among academics. The results are consistent with Table 7.4 with the exception of the number of books listed in the Library of Congress (no longer statistically significant). TED appearances are significantly correlated with higher speaking fees as are NYT Best Selling books. All four proxies remain statistically significant once we control for external influence via Google. Calculating the standardised/beta coefficients indicates the strongest effect for Google pages: for example, three times stronger than having done a TED talk and twice as strong as the number of NYT Best Sellers books.

170

Table 7.5 Academic performance and speaking fees

Dep. Var.: Log(Min. Speaking Fee) (9-22) (23-36) log(non-educational domain Google pages) Coeff. t-stat. Coeff. t-stat. Coeff. t-stat. Publish or Perish Google Scholar Number of papers 7.4e-04** (3.25) -1.4e-04 (-0.54) 0.153*** (5.77) Number of citations 5.2e-06** (2.77) 3.2e-07 (0.17) 0.143*** (6.01) Average number of citations per paper 3.7e-03*** (3.98) 2.8e-03** (3.12) 0.134*** (6.17) h-index 7.1e-03*** (3.91) 1.5e-03 (0.75) 0.135*** (5.35) hI,annual 0.445*** (4.17) 0.18 (1.58) 0.128*** (5.34)

Scopus Total publication count 8.0e-04 (1.58) 4.8e-04 (1.01) 0.143*** (6.56) Total citation count 8.8e-06 (1.56) 4.8e-06 (0.89) 0.143*** (6.54) Average citation count per publication 3.2e-03* (2.59) 1.8e-03 (1.57) 0.142*** (6.15)

Books Number of books on Library of Congress 2.5e-03 (0.86) -3.1e-03 (-1.10) 0.152*** (6.72) Non-fiction book award dummy 0.344 (1.26) 0.189 (0.74) 0.143*** (6.59) NYT Best Sellers (number of books) 0.32*** (4.62) 0.226*** (3.34) 0.128*** (5.82) NYT Best Sellers (number of weeks) 0.011* (2.49) 7.3e-03 (1.69) 0.14*** (6.40)

TED TED dummy variable 0.372** (2.83) 0.234 (1.86) 0.138*** (6.29) TED number of invitations 0.242*** (3.74) 0.159* (2.53) 0.133*** (6.04) Notes: Summary of 28 regressions. N of academics = 314. The symbols *, **, *** represent statistical significance at the 5%, 1% and 0.1% levels, respectively. The control variables (academic involvement, gender, career age, academic field, Nobel Prize and location) were included in the regression but not reported in the table. The first set of regressions (1) does not control for log(non- education web pages) (column 1 & 2) and while (2) does (column 3 & 4). Column 5 & 6 report the coefficient and t-statistics of log(non-educational domain Google pages).

171

Discussion and conclusions

The results of our study indicate that scholars can capitalise on their external prominence in the speakers’ market. The larger the number of web pages that index a particular academic, the higher the minimum speaking fee he or she attracts. Similarly, having been invited to speak at a TED event is positively correlated with speaking fees. In contrast, research performance in terms of publications and citations does not increase speaking fees above and beyond our Internet measure of external prominence. There is a clear distinction between the capitalization of external and internal prominence.

Research evaluations have become a crucial part in the business of science and technology management (Klavans and Boyack 2008) and we observe a phenomenon Johan Bollen describes as a “Cambrian explosion of metrics” (Van Noorden 2010, p. 864). Broader impact criteria may emerge in the future to evaluate the performance of scientists. In particular, criteria that measure a scientist’s social effect could become more important as, for example, government agencies that support fundamental research and education are putting more emphasis on it. The Internet offers significant potential in the measurement of scholarly impact beyond academia. Some methods may be perceived as “quick-and-dirty”, but an overload of information heightens incentives for decision makers to use such instruments that are fast and easily available. In addition, a large set of available tools may increase the incentive of administrators or evaluators to “play an academic version of Moneyball64” (Priem 2013). On the other hand, as Priem (2013) points out, the Web eliminates the artificial distinction between process and product, providing new ways of mapping scholarly contribution: “Suddenly, the rocky plain of ideas once navigated using cairns of citation—is covered in fresh snow. In the Web era, scholarship leaves footprints” (p. 438). Thus, a deeper discussion and exploration of available proxies would be beneficial for academics and beyond.

Future research could look more closely at how the organizational context affects scientists’ external influence. A better understanding of all the different reward structures in academia is crucial, as scientific work and productivity could depend on it. It has been suggested that scientists respond

64 http://en.wikipedia.org/wiki/Moneyball_(film).

173

to the achievement of recognition (Merton 1973). For example, Nederhof (2008) reports results from a natural experiment in the Netherlands where a grassroots ranking (Top 40) led scientists to publish more in these top forty journals. However, such publication stimulus was not connected to an optimization of citation performance.

The willingness to pay for listening to scientists present their research may be taken to reflect the relevance of the science for practical issues. Observing that scholars of more renown in the general public are better paid in the speakers’ market might not strike the reader as surprising. Moving beyond this confirmatory insight, our paper focused on the difference between academic and nonacademic speakers as well as differences between fields. Social scientists in the area of business generate the largest speaking fees compared to other social scientists or natural scientists. In addition, part- time scientists are also more successful. The finding that academic speakers can monetise their knowledge to a larger extent than nonacademic speakers may suggest that there is a considerable link between scholarly research and practical relevance. However, this only holds if speaking fees comprehensively capture the public interest in science. These considerations point out the need to further inquire into the relationship between academic research and practice. The Nobel laureate Rowland (1993) stated in his presidential address to the American Association for the Advancement of Science: “From my own experience, I see that the most serious problems are related to faulty communication about science among the various segments of society, including the scientific segment itself. Each of us is bombarded daily by messages from television, radio, magazines, newspapers, and so on. We live in the midst of massive information flow, but those items connected with science itself are often badly garbled, sometimes with potentially serious negative consequences. The remedy must lie in greater emphasis by all of us increasing both the base level of knowledge of science and communication about science with all levels of society” (p. 1571).

174

Appendix

Figure 7.4 Speaking fees and external impact in sub-fields

175

Table 7.6 Correlation Matrix of internal impact proxies (Google Scholar and Scopus)

Google Scholar Scopus Total Total Avg. h- hIa- Total Total Avg. Papers Cit. Cit. index index Pub. Cit. Cit. Google Scholar Total Papers 1 Total Cit. 0.77 1 Avg. Cit. 0.36 0.64 1 h-index 0.88 0.83 0.55 1 hIa-index 0.58 0.58 0.49 0.75 1 Scopus Total Pub. 0.46 0.37 0.22 0.61 0.50 1 Total Cit. 0.39 0.48 0.33 0.59 0.47 0.82 1 Avg. Cit. 0.36 0.57 0.58 0.53 0.41 0.23 0.45 1

177

Chapter 8 The Inner Quality of an Article: Will Time Tell?

Chan Ho Fai, Malka Guillot, Lionel Page & Benno Torgler

Scientometrics (2015), 104(1), 19-41.

Abstract

In this paper, we assess whether quality survives the test of time in academia by comparing up to 80 years of academic journal article citations from two top journals, Econometrica and the American Economic Review. The research setting under analysis is analogous to a controlled real world experiment in that it involves a homogeneous task (trying to publish in top journals) by individuals with a homogenous job profile (academics) in a specific research environment (economics and econometrics). Comparing articles published concurrently in the same outlet at the same time (same issue) indicates that symbolic capital or power due to institutional affiliation or connection does seem to boost citation success at the beginning, giving those educated at or affiliated with leading universities an initial comparative advantage. Such advantage, however, does not hold in the long run: at a later stage, the publications of other researchers become as or even more successful.

Each period is dominated by a mood, with the result that most men fail to see the tyrant who rules over them. Albert Einstein to Maurice Solovine in 1938 (see Einstein and Infeld, 1938, The Evolution of Physics, p. xxii).

Time is the best censor. Frédérique Chopin (letter to his family, 1846)

How many errors Time has patience for, W. H. Auden (first stanza of Our Bias).

179

Introduction

Does the quality of a scientific contribution survive the test of time? Landes (2003: 144) argues that “[t]ime exposes fads, flash-in-the-pans, and one-time wonders. More controversial is the claim that works that stand time’s test tend to be the most important and influential art of the past.” This paper addresses this important and challenging question. Of course, if the term quality refers to the importance of a scientific contribution, it is difficult to provide a definitive and quantifiable answer. Mazlish (1982) proposes a distinction between an “inside” dimension of scientific quality shaped by the scientific profession’s own assessment of scientific work and an “outside” quality decided by social evaluation. Here, we rely on citations to study the academic 65 environment and thus concentrate on the inner dimension. Empirically, this use of citation counts as a measure for quality is very convenient because of ready availability and the objective measurement provided. There is also substantial evidence justifying its use as a (rough) quality measured. For instance, citations are highly correlated with peer ratings of eminence or perceived scientific significance (Albert 1975; Lawani 1986). The seminal work of Merton (1957) even suggests that a paper’s quality can be appraised by its citation counts.

In Merton’s (1973) theory, a citation has two functions arising from the normative structure of science. First, authors use citations to highlight the work that has influenced their research and to indicate further readings that might be of interest to the reader, which can thus be seen as a cognitive function of citations. Second, scholars use citations to pay an intellectual debt by helping the authors cited to become better known. Thus, citations are a form of recognition. Obviously, however, the likelihood of being cited (and thus citation counts) is influenced by many factors (see, e.g., Bornmann and Daniel 2008), including those related to the timing, field of research, journal, article, and author/readership.

Citations, then, represent a complex phenomenon that cannot be explained simply by the intellectual content of an article. As Stigler et al. (1995: 344) point out, a network of citations is the “product of a complex

65 For a quantitative analysis of the outside dimension, see Chan, Frey, Gallus, Schaffner, Torgler and Whyte (2013, 2014).

180 combination of factors, ranking from scientific influence and social contact to an element of pure chance in the timing of publication of accepted papers.” Thus, social context also matters because scientific knowledge is generated through a social process (Latour and Wooglar 1979). As a result, citations are not only used to acknowledge intellectual debt but also as, for example, rhetorical tools. That is, citing certain authors provides support for a paper and persuades the scientific community of the validity of the findings (Gilbert 1977). On the other hand, citations are also subject to bias. For example, “hat- tipping” citations may be introduced to please authors that could be potential referees, to demonstrate that the relevant literature has been read, or in the hope that cited authors will reciprocate in the future (Mayer 2004: 624). In other words, there is contamination through manipulation (Merton 1973). This possibility that distortion may go hand in hand with an unequal distribution of citations has led to the development of various theoretical concepts. Merton, for instance, speaks of “the Matthew effect” (Merton 1968, 1988), referring to a phenomenon in which success breeds success.66 In this dynamic, authors not only profit from their own reputation and that of their institutions and network but also from symbolic capital or power (Bourdieu 1989; Putnam 2009), defined here as power that signals academic legitimacy and thus also academic reputation, status, and authority, thereby promoting comparative advantage. The decision to read a paper may thus be based on whether it is written by someone from a top-tier university or originates from a lesser known department (even when published in a top-tier journal). In that case, authors could benefit from the fame of the institution at which they completed their doctorate or, more visible to the reader, at which they were working at the time of paper publication. Skewed citation distribution could thus lead to a process of advantage or disadvantage accumulation. This question of whether author characteristics and the scope of their work might alter the frequency and adequacy of citation counts through a reputation effect is crucial to our use of them as a quality measure.

66 For more recent studies, see Watts and Gilbert (2011) for an agent-based simulation, and Azoulay et al. (2014) and Chan, Frey, Gallus and Torgler (2014a) for the citation patterns of papers published before the bestowal of an award. Although both these latter construct synthetic counterfactuals with the same preaward citation structure, Azoulay et al. (2014) observe only a small citation boost over a short period because of the award, while Chan, Frey, Gallus and Torgler (2014b) observe a very large and long-lasting effect.

181

Many articles also stress the unequal nature of productivity in science, a reality first revealed by Lotka (1926), who showed that half of all 67 papers were published by only 6 % of publishing scientists. Since then, a large body of literature has specifically explored the dynamics of citations or, in particular, the citation trajectories of papers (see, e.g., Price 1976; Chubin et al. 1984; Aversa 1985; Garfield 1989, 1990; Redner 1998, 2005; Glänzel et al. 2003; Mingers 2008; Levitt and Thelwall 2008; Wallace et al. 2009; Hsu and Huang 2011; Roth et al. 2012; Eom and Fortunato 2011; Ohba and Nakao 2012; Hjørland 2013; Sangwal 2013; Wang et al. 2013; Watts and Gilbert 2011; Bjork et al. 2014; Ponomarev et al. 2014).

In this paper, we are interested in exploring whether articles have an inner quality as opposed to various types of bias that may manifest particularly in the early years after publication. We therefore ask whether effects beyond quality become ever less important over time or are cumulative. We also examine whether potential differences related to social contacts, professional networks, or scientific influence (i.e., through 68 institutional and doctoral affiliation) disappear over time. In other words, does time reveal the inner quality of an article? Peter Carruthers, a former leader of the Los Alamos Theoretical Division, argues that “…the quality [of scientific work] survives miraculously, despite all the human foibles that are translated into the way science is done. That’s largely due to the experimentalists, I suppose. Somehow science is self-correcting. Even though credit often is assigned unfairly, the actual evolution goes on, you sort out the better ideas from the junk, and occasionally there are major insights” (Simmons and West 1981: 139).

Obviously, this question of whether quality survives the test of time is interesting even beyond the academic environment. However, the academic context is analogous to an experimental setting in that it features a homogeneous task [trying to publish in top-tier journals, namely

67 See, for example, Price (1963), Coles (1970), Allison and Stewart (1974), Allison (1980), Redner (1998, 2005). For one of the journals that we analyse, the American Economic Review, 80 % of the citations received within the 1911–2011 period are from 20 % of the articles (Torgler and Piatti 2013). 68Admittedly, authors who studied at or work at a leading university may not only have better connections or an ability to influence the subject/topic of publications but may also be able to amass substantial experience, gather feedback and inspiration, and be exposed to the type of training that may be used to develop research that increases the inner quality of a paper.

182

Econometrica and American Economic Review (AER)] by individuals with a homogenous job profile (academician) in a specific research environment (economics and econometrics). In addition, comparing papers published at the same time (same issue) provides a comparable group of papers (articles judged as worth publishing by editors and referees). Focusing on a specific journal also allows better control of the channel through which the articles were published.

Data

Our citation count data are drawn from almost 80 years of articles. The journals we explore, namely AER and Econometrica, are recognised as among the best economics journals (Kalaitzidakis et al. 2003, 2011; Wall 2009; Engemann and Wall 2009; Kodrzycki and Yu 2006; Axarloglou and 69 Theoharakis 2003). These two journals do, however, attract slightly different submissions: whereas AER is a more general economic journal, Econometrica is more theoretically driven. Such a difference is useful in that it allows us to test the robustness of our results and increase the range of their validity. For Econometrica we have collected a larger sample. To this end, our primary focus will be on Econometrica, with AER used for robustness tests. In particular, Econometrica is more specialised and thus less driven by such biases as the size of the subfield.

All citation data were generated through the ISI Web of Knowledge provided by Thomson Reuters. The first and larger sample comprises 3247 papers published in Econometrica between 1933 and 2010. The second sample consists of 409 papers published in issues 1, 3, 4, and 5 of the AER 70 between 1984 and 1988. To increase sample homogeneity, we focus on original contributions, excluding all post-publication papers like replies, comments, or corrections whose impact in terms of explanatory variables may not be the same as that of full papers.

69 Kalaitzidakis et al. (2011) rank Econometrica and AER second and first in economics journals, respectively. The two are also ranked third and fourth, respectively, in the “ambition-adjusted journal ranking” devised by Engemann and Wall (2009). 70 We exclude the Papers and Proceedings.

183

To test for the influence of institutional environment, we compare the citation performance of articles by authors from the world’s top 10 and top 20 universities against the performance of papers whose authors are unaffiliated with such institutions using the ranking developed by Amir and Knauff (2008). Based on this dichotomy, we develop three categories: (1) none of the authors belong to such a university; (2) all of the authors belong to such a university; (3) at least one author but not all authors belong to such a university (mixed category). In addition, we include a variable for whether author doctorates were completed at a university ranked in the top 10 or top 20 positions. Amir and Knauff’s (2008) ranking is based on the strength of its Ph.D. program (see Appendix Table 8.4). The criterion for this ranking is a department’s ability to place doctoral graduates in top-level economics departments or business schools. The authors themselves describe the methodology as follows: “For an n-department sample, the idea is to derive an endogenous relative valuation of each department by specifying a system of n equations wherein the value of department i is a weighted average of the values of all other departments, with the jth weight being the number of placements department i has made in department j. Thus the value of each placement is given by the score of the employing department, which is itself simultaneously determined in the underlying fixed point relationship. The final score of a department is then simply the sum of all the values of its individual placements” (Amir and Knauff 2008: 185). Based on data collected from the Web in April 2006, these authors claim that faculty hires might be a more reliable and stable indicator of influence than journal citations.

The author affiliation at the time of publication is listed on the article itself. When authors report two or more work affiliations, we take the affiliation with the highest institutional ranking. To locate the institutions at which authors earned their doctoral degrees, we search for CVs and check for a thesis/dissertation record under the author’s name on digital dissertation 71 archives such as ProQuest Dissertations & Theses Global, as well as any dissertation databases available from the top 20 universities. In the Econometrica sample, the authors of 635 out of 3247 articles (19.6 %) are all from a top 10 university versus at least one author (but not all authors) of 289 articles (8.9 %). Of all 409 articles in the AER sample, 90 articles (22 %) are

71 www.proquest.com/products-services/pqdtglobal.html.

184 by authors with a top 10 university affiliation at the time of publication versus 32 articles (7.8 %) by at least one author (but not all authors). Likewise, the authors of 1225 articles (37.7 %) all obtained a doctorate at a top 10 university versus at least one author (but not all authors) of 531 articles (16.4 %). For AER, all the authors of 206 articles (50.4 %) earned their doctorates at a top 10 university versus at least one author (but not all authors) of 68 articles (16.6 %).

Using citation as a measure of article quality, however, is not unproblematic. For example, it is evident that papers with multiple authors attract more citations than single authored papers (Ductor 2014). Thus, simply comparing the citation differences for single authored works with those for multiple authored works could lead to biased results, especially given that our mixed category consists of only multiple authored papers while the other two categories (only top university and only non-top university) have a combination of single and multiple authored contributions. We therefore normalise the raw citation count by dividing it by the square root value of the number of co-authors. Another source of possible bias is that citations often follow a power law distribution (Gupta et al. 2005; Redner 2005), meaning that results could be driven by a handful of frequently cited papers. Thus, following Huang (Huang 2015), for each year we rank the papers based on the yearly citations received relative to the yearly citations received by other papers. Using the citation counts normalised by number of co-authors, we define the citation rank of paper i in year t as

, + 1 Rank = × 100 cit

Descriptive analysis

To assess the citation differences between articles written by authors of top and non-top university based on the ranking classification, we calculate

185 citation ranks of papers in each category published in the same issue. By doing so, we avoid the problem resulting from comparing the citation trajectories of papers from different cohorts. Another advantage is that we can hold the standards for paper acceptance constant, which allows comparison of similar quality articles (i.e., those judged worthy by the same managing editor, co-editors, and editorial board members). Drawing from 385 issues of Econometrica and 20 issues of AER, we set the citation performance of articles with no author from a top 10 university as the base line (horizontal line at value zero) and then compare it against the citation differences between articles having all authors from a top 10 university (blue) and those having at least one author from a top 10 university (red). We depict these citation rank differences in Figure 8.1 on a yearly basis.

186

Figure 8.1 Citation rank difference over time for authors belonging or not to a top ten university

Note: The IQRs (interquartile ranges) represent the ranges between observations at the 25th and 75th percentiles. IQRs for the Mixed category are offset by +0.5 years for better visualization.

In Table 8.1 and Table 8.2 we report pairwise t tests exploring the statistical significance of these differences over time. For both journals, we observe a rapid increase in citation rank difference within 5 years of publication, which suggests that immediately after publication, articles by authors from a top 10 university attract more citations than articles in the same issue by authors from a non-top 10 university. For example, 5 years after the publication year, a Top 10 Uni and a Mixed article in Econometrica are ranked, on average, 6.99 and 9.12 higher than the baseline in terms of

187

cumulative citations, differences that are statistically significant at the 1 % level. On the other hand, the average citation rank differences for Top 10 Uni and Mixed article in AER are 13.09 and 9.16, respectively (statistically significant at the 1 and 10 % levels, respectively). Admittedly, such results may be driven by reputation or more generally by symbolic capital or power. Nevertheless, after year 5, the rate at which the mean citation rank difference increases begins slowing down, in particular for Top 10 Uni in Econometrica and Top 10 Uni and Mixed in AER. This finding suggests no Matthew effect and a deterioration of the importance of symbolic capital. The t test results even show that 40 years after publication, the citation rank difference (based on cumulative citation) for Top 10 Uni in Econometrica has become insignificantly different from 0 (at the 10 % level), indicating a convergence in the two groups’ citation patterns. In AER, this difference remains significant until 25 years after publication, but the statistical significance of the difference in the cumulative citation rank drops after 5 years.72 The t test results for top 20 universities show a similar adjustment process. Authors based on doctoral university rather than current affiliation show smaller differences between those with and without a top 10 or top 20 doctorate but a slower adjustment process.73

Surprisingly, 30 years after publication, the mean citation rank difference for Mixed articles in Econometrica suddenly increases after remaining relatively flat for three decades. We therefore take a closer look by splitting the sample by decade (see Figure 8.2). We observe that the patterns of citation rank difference in both categories remain flat over time (relative to the baseline) for articles published in most decades except the 1960s and 1940s, during which the citation rank difference for Mixed continues to increase. The results for the 1940s, however, should be treated with caution as the sample size decreases in the later years while there are only a limited number of Mixed papers in the early years. For instance, prior to 1950, only three issues (out of 67) contain at least one article classified in the Mixed category. We obtain a similar result for the same citation rank difference based on top 20 universities (see Appendix Figure 8.4 and Figure 8.5). However, the interquartile ranges (IQR) in Figure 8.4 and Figure 8.5 are wider

72 Here, the sample size is reduced due to a lack of observations. 73 Results are available from the authors upon request.

188 than those in Figure 8.1 and Figure 8.2 (top ten universities) because the IQR contain a greater number of low values for citation rank difference. The citation rank differences based on author doctorates shows a similar pattern (see Appendix Figure 8.6 and Figure 8.7); however, the IQRs are substantially larger and lower. Such a result is in line with our expectations based on the fact that information on author doctorate is less visible than current institutional affiliation.

Table 8.1 Mean citation rank difference in Econometrica, by year since publication

Top 10 university vs. non top 10 Mixed vs. non top 10 university university Year since # Citation Citation rank # Citation Citation rank publication issues rank difference issues rank difference difference (cumulative difference (cumulative citation) citation) 0 305 2.1 0.34 187 6.13*** 2.23*** 1 300 8.16*** 3.97*** 181 14.61*** 7.79*** 2 296 6.43*** 5.86*** 176 6.96*** 7.06*** 3 293 9.03*** 7.30*** 172 10.26*** 7.91*** 4 288 6.85*** 7.36*** 167 13.16*** 8.68*** 5 283 5.97*** 6.99*** 164 13.29*** 9.12*** 6 278 11.49*** 7.60*** 159 11.64*** 9.23*** 7 272 9.32*** 7.55*** 154 12.89*** 9.51*** 8 266 11.01*** 7.87*** 150 10.44*** 9.35*** 9 262 8.99*** 7.71*** 146 14.41*** 9.61*** 10 258 10.82*** 7.90*** 141 16.13*** 10.27*** 11 254 9.51*** 7.92*** 138 13.42*** 10.50*** 12 250 10.58*** 7.79*** 132 16.45*** 10.97*** 13 247 8.52*** 7.59*** 128 15.49*** 10.86*** 14 242 12.20*** 7.72*** 126 12.56*** 10.64*** 15 237 10.95*** 8.09*** 121 14.15*** 10.29*** 16 233 11.46*** 7.90*** 116 16.01*** 10.31*** 17 229 10.72*** 7.77*** 112 16.91*** 10.56*** 18 224 9.79*** 7.53*** 108 18.99*** 10.75*** 19 219 9.60*** 7.69*** 105 19.02*** 11.06*** 20 213 10.08*** 7.95*** 101 19.40*** 11.50*** 21 208 9.12*** 7.76*** 96 14.55*** 11.30*** 22 203 9.57*** 7.43*** 91 11.52*** 10.88*** 23 197 7.97*** 7.19*** 86 16.60*** 10.65*** 24 191 6.35*** 7.01*** 81 21.61*** 12.53*** 25 185 10.62*** 7.14*** 77 16.45*** 12.30*** 26 179 7.69*** 7.05*** 74 12.25** 11.89*** 27 174 6.92*** 6.60*** 69 12.87*** 11.70*** 28 169 5.16** 6.66*** 63 11.13** 10.74*** 29 163 6.22** 6.68*** 60 9.76** 11.22*** 30 157 6.50** 6.78*** 56 13.15** 10.49*** 31 150 6.39** 6.72*** 52 10.19** 10.98*** 32 144 6.95*** 6.87*** 48 7.57 12.66*** 33 138 6.00** 6.30*** 45 15.10** 14.35*** 34 130 4.99* 6.00*** 40 15.35*** 14.96*** 35 127 5.54** 5.81** 37 21.51*** 16.63*** 36 123 4.83* 5.82** 34 10.87* 18.19*** 37 117 9.04*** 5.49** 31 14.41** 17.91*** 38 111 6.31** 4.12* 29 12.86* 17.34*** 39 105 2.57 4.08* 28 14.76* 18.57***

189

Table 8.1 continued

Top 10 university vs. non top 10 Mixed vs. non top 10 university university Year since # Citation Citation rank # issues Citation Citation rank publication issues rank difference rank difference difference (cumulative difference (cumulative citation) citation) 40 100 2.66 3.47 26 5 20.49*** 41 96 6.53* 3.66 22 23.57*** 23.15*** 42 92 1.61 3.53 22 31.77*** 23.33*** 43 90 5.25 3.86 21 23.42*** 25.10*** 44 87 1.19 4.72* 19 33.12*** 28.09*** 45 83 4.87 5.08* 16 12.95 28.71*** 46 79 0.84 5.61* 16 26.79** 28.72*** 47 77 4.23 5.31* 15 25.82** 29.85*** 48 74 6.06 5.60* 14 30.65** 33.42*** 49 71 3.9 5.74* 14 24.59* 33.33*** 50 68 5.15 5.45 12 27.07** 35.64*** 51 64 1.71 4.73 10 34.00** 37.72*** 52 60 2.06 3.97 9 25.93 37.05*** 53 56 1.05 3.16 9 17.91 36.88*** 54 53 3.48 1.74 9 46.44*** 37.05*** 55 51 0.63 1.98 7 41.60** 42.36** 56 48 2.89 1.19 6 35.75 40.56** 57 45 2.95 1.68 5 32.14 32.60* 58 43 -5.51 0.29 4 21.16 28.26 59 39 -0.23 0.83 4 37.46 28.81 60 37 3.39 0.94 3 20.14 29.55 61 34 0.15 1.09 3 -4.44 29.55 62 34 0.57 1.23 2 34.42 49.23 63 33 0.3 1.49 2 55.29 49.37 64 31 4.52 0.27 1 67.13 71.27 65 29 1.39 0.61 1 87.79 71.92 66 26 0.27 2.95 1 87.23 72.53 67 25 -4.8 1.7 68 24 0.36 2.17 69 21 2.99 -1.79 70 19 -2.92 -2.56 71 16 -5.79 -3.15 72 13 -7.67 -1.12 73 11 -18.22 -0.91 74 8 20.19 13.75 75 6 6.06 3.64 76 4 12.08 12.79 77 3 8.5 25.12 Notes: One sample t-test on the significance of citation differences not equal to 0. *, **, *** represent statistical significance at the 10%, 5% and 1% levels, respectively.

190

Table 8.2 Mean citation rank difference in AER, by year since publication

Top 10 university vs. non top 10 university Mixed vs. non top 10 university Year since # issues Citation Citation rank # Citation Citation rank publication rank difference issues rank difference difference (cumulative difference (cumulative citation) citation) 0 20 1.36 1.16 16 2.47 2.88 1 20 12.24** 13.10*** 16 22.29*** 16.86** 2 20 14.33*** 14.01*** 16 15.81** 14.49** 3 20 14.04*** 13.16*** 16 14.43** 12.37** 4 20 19.49*** 14.00*** 16 9.03 10.68* 5 20 12.46*** 13.09*** 16 10.91* 9.16* 6 20 10.36* 12.74*** 16 4.76 7.84 7 20 10.87* 12.22*** 16 8.03 7.07 8 20 7.04 11.64*** 16 12 7.22 9 20 10.41* 11.25*** 16 7.14 6.44 10 20 8.53 11.06*** 16 13.02 7.39 11 20 13.87*** 10.99*** 16 15.65** 7.68 12 20 11.90** 11.09*** 16 14.64* 8.3 13 20 14.12** 11.22*** 16 9.99 7.95 14 20 15.20*** 11.38*** 16 15.12** 7.97 15 20 19.31*** 11.71*** 16 13.72* 8.08 16 20 13.89*** 11.78*** 16 15.09** 8.17 17 20 13.86*** 11.95*** 16 17.45* 8.32 18 20 7.77 11.69*** 16 11.14 8.11 19 20 14.04*** 11.79*** 16 12.81 8.23 20 20 18.87*** 11.80*** 16 12.49* 8.35 21 20 19.02*** 11.92*** 16 11.29 8.39 22 20 6.06 11.68*** 16 18.71*** 8.44 23 20 14.93*** 13.52*** 16 10.56 9.13 24 16 12.10** 13.05*** 15 13.28 8.49 25 12 -3.54 9.82* 11 12.43 3.28 26 8 21.09 4.82 8 4.3 6.21 27 4 1.36 1.16 4 2.47 2.88

191

Figure 8.2 Citation rank difference over time in Econometrica, by decade of publication

Note: The 1930s decade includes articles published from 1933 to 1939, but this category contains no authors from the mixed category. Articles published in 2010 are also included in the 2000s

This descriptive analysis highlights the influence of several factors on the pattern of mean citations over time. The working environment at the time of publication shows a skewed distribution of citations in favour of the top environments in the early years after publication. It thus seems to be a potential advantage that can increase an article’s citations relative to the intrinsic quality of the paper itself. Nevertheless, although the descriptive

192 facts help us establish a correlation between the different variables studies (raw effect), a multivariate analysis is needed to uncover causality.

Multivariate analysis

To estimate the effect that an author’s affiliation with or doctorate from a particular category of university (top ten versus non-top ten) exerts on citation ranks, we model the citation rank (based on citation count adjusted by number of authors) of paper i in year t using a random effect generalised least squares (GLS) model (Table 8.3). Since the aim of our study is to analyse the influence of time on citation rank, we include a time variable, number of years since publication, as an explanatory variable. This inclusion puts all articles on equal footing with respect to the citation count in 1 year. We also include two dummy categories for the author’s university affiliation at the time of publication and another two for doctoral program (i.e., top ten institutions and top ten Ph.D. institutions). Since we want to estimate whether the effect of the university environment depends on the number of years since publication, we also include interaction terms between this variable and the two university variables, which allows analysis of the differential effects of an additional year between the categories. As best fit for the time effect, we identify a quadratic relation for AER but a cubic relation for Econometrica. As control variables, we include paper length (length); proportion of male authors (share male), and mean academic age of the authors, defined as the year of publication minus the year the doctorate was obtained (academic age). To ensure closeness to the pairwise comparison in the descriptive analysis, we further include dummy variables for each issue to hold them constant in the estimates.

Table 8.3 presents the estimates for the Econometrica papers in columns (1) and (2) and those for AER in columns (3) and (4) based on top ten affiliations. To better depict the quantitative effects, we show the estimated adjusted means for all years since publication (in 1-year increments) for the two top ten groups in relation to the baseline (see Figure 8.3). This arrangement allows us to test the equality of these groups to the reference group (papers by authors of non-top ten institutions). The quadratic relation in AER indicates an increase in the difference to the baseline over time, reaching its strongest point in year 13 and decreasing thereafter. The

193 cubic relation, however, shows that after a while this difference increases yet again. Nevertheless, the Mixed author results should be treated with caution because the number of such papers is limited early in the history of Econometrica (Table 8.1). Thus, we must warn the reader when interpreting the exploding tail of the curves as it may be caused by the decreasing number of observations during the early years or the cubic (and quadratic) polynomial. Figure 8.3 also shows that the patterns for doctoral affiliation are very similar to the institutional one. Moreover, results extending it to top 20 places are also comparable (see Figure 8.8).

Table 8.3 Results of random-effects GLS regression models (top 10 university and Ph.D.)

Econometrica AER Variables (1) (2) (3) (4) Years since publication 0.329*** 0.174 1.668*** 1.229*** (YP) (0.078) (0.090) (0.243) (0.347) Years since publication2 -0.028*** -0.022*** -0.065*** -0.045*** (YPSQ) (0.003) (0.003) (0.009) (0.013) Years since publication3 3.6e-04*** 3.0e-04*** (YP3) (3.2e-05) (3.5e-05) All top 10 uni 5.215*** 3.577 (1.191) (2.869) Mixed top 10 uni -0.620 2.975 (1.758) (4.572) All top 10 uni*YP 0.588*** 0.769 (0.176) (0.480) Mixed top 10 uni*YP 2.515*** 0.474 (0.352) (0.744) All top 10 uni*YPSQ -0.021** -0.029 (0.007) (0.017) Mixed top 10 uni*YPSQ -0.120*** -0.019 (0.018) (0.026) All top 10 uni*YP3 2.2e-04** (7.9e-05) Mixed top 10 uni*YP3 1.4e-03*** (2.5e-04) All top 10 PhD 1.778 -0.776 (1.081) (2.556) Mixed top 10 PhD -4.602** -3.472 (1.445) (3.643) All top 10 PhD*YP 0.544*** 0.881* (0.151) (0.443) Mixed top 10 PhD*YP 1.942*** 1.222 (0.287) (0.640) All top 10 PhD*YPSQ -0.022*** -0.038* (0.006) (0.016)

194

Table 8.3 continued

Econometrica AER Variables (1) (2) (3) (4) Mixed top 10 -0.091*** -0.053* PhD*YPSQ (0.015) (0.022) All top 10 PhD*YP3 2.2e-04** (6.7e-05) Mixed top 10 PhD*YP3 1.1e-03*** (2.1e-04) Article length 0.658*** 0.692*** 2.052*** 2.152*** (0.050) (0.050) (0.230) (0.230) Share male -3.823* -3.639* 1.743 1.816 (1.765) (1.815) (6.029) (6.034) Academic age -0.118** -0.095* 0.941 1.227 (0.045) (0.046) (1.447) (1.448) Issue fixed effect YES YES YES YES Observations 93423 93423 10177 10177 Number of articles 2960 2960 407 407 R2 0.473 0.462 0.263 0.254 Notes: The paper type reference group is no author affiliated with a top 10 university in models (1) and (3), and no author completed a doctorate in a top 10 university in models (2) and (4). Standard errors are in parentheses. *, **, *** represent statistical significance at the 10%, 5%, and 1% levels, respectively.

Figure 8.3 Contrasts of predictive margins (by top ten university)

195

Among the other control variables, only paper length seems to have a consistently and significantly positive robust effect on citation rank, which echoes Hudson’s (2007) finding for AER and the Economic Journal. The effect of academic age is statistically significant in Econometrica but not in AER, suggesting that younger scientists are more successful in the former than the latter. The effect of gender, however, is unclear: the proportion of males even has a significantly negative effect in the Econometrica regression that is inconsistent with studies addressing this question. Based on earlier research, men should receive significantly more citations than women. Not only do Cole and Singer (Cole and Singer 1991) demonstrate that being a man has a positive effect on the number of citations received, but Stack (2004) shows that the research productivity of women is lower, even when the number of young children is controlled for. Moreover, Baldi (1998: 842), after modelling a citation as a dyadic relationship between a cited and citing author, concludes that “scientists are significantly less likely to cite articles written by female authors.”

As a whole, the multivariate analysis confirms that the work and educational environment influences the number of citations received, particularly during the first few years after publication. This finding seems to indicate that symbolic capital or power matters in this time period. The interaction of this variable with the time component confirms that this early advantage tends to stabilise, although it also shows a catching-up effect in some cases. This effect is particularly noticeable for the AER sample in which the interaction terms between years since publication and its squared term and the categorical variable for publication environment indicates that the negative effect of the squared years overwhelms the positive effect after 13 years. Thus, after an advantageous start, articles that profit from symbolic capital are caught up with in terms of citation count, implying that the inner quality of an article is revealed over time. Econometrica also shows adjustments that support this argument. In particular, we observe stabilization in the pure top 10 or top 20 category relative to the baseline and even a decrease for the mixed group up to year 42, after which the difference from the baseline is even below zero. For a small sample of articles, however, the relative difference in citation success increases again over time in later years.

196

Conclusions

The interesting question of whether the quality of a scientific contribution survives the test of time has as yet not been intensively empirically explored. In this paper, by comparing articles published at the same time in the same outlet (i.e., the same issue of a volume), we find evidence of potential biases due to institutional affiliation or connection, which suggests that authors profit from the symbolic capital or power of a top university. Such a comparative advantage disappears over time, however, through stabilization of the relative difference and, except for a few articles, even decreases over time. Interpreting this result in light of our hypothesis, we conclude that the inner quality of the papers published in these top journals is revealed over time.

Admittedly, this analysis has certain limitations, especially in terms of the fundamental assumptions that are crucial to our model. First, we assume that papers published in the same journal are roughly of the same (perceived) quality. We also assume that categorization of the authors (by top 10 or top 20 universities versus others) is a valid proxy for the type of research environment. Such an assumption might be justifiable prima facie, but other elements (e.g., author reputation) may also be relevant. We were also unable to distinguish self-citations, a distinction that might improve the relevance of the results. For example, Johnston et al. (2013) note that although some argue that self-citation is self-serving, others believe it is central to the progression of scientific communication. There is also evidence that self-citation has no significant quantitative effect on the total number of citations. A further drawback is the possibility of selection bias in the original publication process as a result of editor or referee predilections.

With respect to the use of citations as our variable of interest, a consideration of the context in which the citations are made (e.g., the quality of the journal in which the article is cited) might improve analytic quality. However, not only would it be difficult in this present analysis to account for context over the extremely long investigatory period, but such deconstruction of citation incidence is still in its infancy and thus lacks a developed theoretical framework. Our analysis thus makes a contribution by helping lay the groundwork for this conceptual development.

197

Appendix

Figure 8.4 Citation rank difference over time for authors belonging or not to a top 20 university

198

Figure 8.5 Citation rank difference over time, by decade of publication (top 20 university)

199

Figure 8.6 Citation rank difference over time for authors obtaining a Ph.D. in a top ten university

200

Figure 8.7 Citation rank difference over time, by decade of publication (top ten Ph.D.)

201

Figure 8.8 Contrasts of predictive margins (by top 20 university and Ph.D.)

202

Table 8.4 Institutional Ranking

Rank University 1 MIT 2 Harvard University 3 Stanford University 4 Princeton University 5 University of Chicago 6 Yale University 7 University of California, Berkeley 8 Oxford University 9 University of Minnesota 10 Northwestern University 11 London School of Economics 12 University of Pennsylvania 13 Carnegie Mellon University 14 University of Rochester 15 University of California, Los Angeles 16 University of Wisconsin 17 University of Michigan 18 Duke University 19 Cambridge University 20 Columbia University Source: Amir and Knauff (2008, p. 188).

203

Table 8.5 Results of random-effects GLS regression models (top 20 university and PhD)

Econometrica AER Independent Variables (1) (2) (3) (4) 0.242** 0.547*** 1.556*** 0.509 Years since publication (YP) (0.087) (0.071) (0.279) (0.510) -0.025*** -0.037*** -0.062*** -0.023 Years since publication2 (YPSQ) (0.003) (0.003) (0.010) (0.018) 3.3e- 4.5e- Years since publication3 (YP3) 04*** 04*** (3.4e-05) (3.1e-05) All top 20 uni 5.774*** 4.599 (1.055) (2.608) Mixed top 20 uni -1.806 5.733 (1.650) (4.279) All top 20 uni*YP 0.526*** 0.772 (0.148) (0.428) Mixed top 20 uni*YP 2.669*** 0.442 (0.335) (0.670) All top 20 uni*YPSQ -0.019** -0.027 (0.006) (0.015) Mixed top 20 uni*YPSQ -0.129*** -0.017 (0.017) (0.023) All top 20 uni*YP3 1.8e-04** (6.5e-05) Mixed top 20 uni*YP3 0.002*** (2.4e-04) All top 20 PhD 8.644*** -1.345 (2.605) (3.052) Mixed top 20 PhD 1.977 -3.482 (2.871) (4.575) All top 20 PhD*YP 0.009 1.594** (0.388) (0.559) Mixed top 20 PhD*YP 2.035*** 1.933* (0.577) (0.794) All top 20 PhD*YPSQ -0.004 -0.058** (0.016) (0.020) Mixed top 20 PhD*YPSQ -0.099** -0.073** (0.035) (0.027) All top 20 PhD*YP3 5.5e-05 (1.6e-04) Mixed top 20 PhD*YP3 0.001* (5.7e-04) Article length 0.657*** 0.687*** 1.990*** 2.086*** (0.049) (0.049) (0.237) (0.226) Share male -4.003* -3.545* 0.071 1.011 (1.800) (1.798) (6.155) (5.744) Academic age -0.104* -0.109* 1.313 1.320 (0.046) (0.044) (1.423) (1.455) Issue fixed effect YES YES YES YES Observations 93423 93423 10177 10177 Number of articles 2960 2960 407 407 R-square 0.475 0.463 0.274 0.260 Notes: The paper type reference group is no author affiliated with a top 20 university in models (1) and (3), and no author completed a doctorate in a top 10 university in models (2) and (4). Standard errors are in parentheses. *, **, *** represent statistical significance at the 10%, 5%, and 1% levels, respectively.

204

Chapter 9 Do Great Minds Appear in Batches?

Chan Ho Fai & Benno Torgler

Scientometrics (2015), 104(1), 19-41.

Abstract

Despite much scholarly fascination with the question of whether great minds appear in cycles, together with some empirical evidence that historical cycles exist, prior studies mostly disregard the “great minds” hypothesis as it relates to scientists. Rather, researchers assume a linear relation based on the argument that science is allied with the development of technology. To probe this issue further, this study uses a ranking of over 5600 scientists based on number of appearances in Google Books over a period of 200 years (1800– 2000). The results point to several peak periods, particularly for scientists born in the 1850–1859, 1897–1906, or 1900–1909 periods, suggesting overall cycles of around 8 years and a positive trend in distinction that lasts around 100 years. Nevertheless, a non-parametric test to determine whether randomness can be rejected indicates that nonrandomness is less apparent, although once we analyse the greatest minds overall, rejection is more likely.

Sporadic great men come everywhere. But for a community to get vibrating through and through with intensely active life, many geniuses coming together and in rapid succession are required. This is why great epochs are so rare – why the sudden bloom of a Greece, an early Rome, a Renaissance, is such a mystery. Blow must follow blow so fast that no cooling can occur in the intervals. James (1880, p. 453)

Introduction

Is there in fact a tendency for an extraordinary number of great minds or geniuses to emerge within a short time span? In other words, do great minds appear in cycles like those that occur in many other areas of life? James

205

(1880), for example, refers to “different cycles of operation in nature” (p. 443), while Schumpeter (1934/1983) stresses that “in their special way both the rise and the fall of families and firms are much more characteristic of the capitalist economic system, of its culture and its results, than any of the things that can be observed in a society which is stationary in the sense that its processes reproduce themselves at a constant rate” (p. 255). Not only have scholars long argued for the existence of historical periods characterised by a large number of gifted people (see, e.g., Bethune in 1837, cited in Becker (1995)), but theoretical particle physicist and Japan’s first Nobel laureate Hideki Yukawa (1907–1981) was certain that great minds appear in batches. As evidence, he refers to the remarkable number of geniuses that emerged during the seventeenth century, including Bacon, Galileo, Kepler, Descartes, Newton, and Leibniz. The beginning of the twentieth century was another such period, populated by thinkers like Planck, Einstein, Rutherford, de Broglie, Born, Heisenberg, Bohr, Schrödinger, and Dirac: “It is usual, it seems, for geniuses to appear in batches; on the other hand, there are also periods during which the appearance of geniuses is very rare. There must be some reason for this, I feel, other than coincidence” (Yukawa 1978, p. 127).

People have also been fascinated by the question of whether it is possible to “manufacture” geniuses, with the names John Stuart Mill, Norbert Wiener, or William Sidis cited frequently in discussions of parental attempts to educate (see, e.g., Howe 1999). Society also appears interested in why some thinkers become famous while others are buried in obscurity: “Almost no one can claim to be an ‘educated layman’ without being able to recognise the name of the philosopher Descartes, yet how many college graduates have even heard of Descartes’ contemporary thinker Henri de Roy?” (Simonton

74 1976a, p. 630). Many years ago, James (1880) suggested that “[t]he causes of production of great men lie in a sphere wholly inaccessible to the social philosopher. He must simply accept geniuses as data, just as Darwin accepts his spontaneous variations. For him, as for Darwin, the only problem is, these date [sic] being given, How [sic] does the environment affect them, and how do they affect the environment?” (p. 445).

74 For an overview of the studies related to society and creativity, see Gowan and Olson (1979).

206

Meanwhile, several papers have explored the question of why such geniuses emerge. Naroll et al. (1971), for instance, offer descriptive evidence that the more politically fragmented a civilization, the higher its level of creativity. Perhaps the most significant contribution, however, and one that provides extensive empirical evidence on this issue, is that of Simonton (1975a, b, 1976a, b, c, 1978, 1980, 1984). In particular, Simonton stresses that sociocultural conditions and events such as zeitgeist, political fragmentation, war, civil disturbances, and political instability affect the development of creative potential in a youthful genius, which can mean the difference between Golden Ages and periods of drought with respect to creative giants. Warfare, for example, has an adverse impact on creativity because it causes stress and anxiety and diverts the investment of time, money, and labour away from imaginative pursuits (Simonton 1975a). However, the type of war is also important. Wars fought close to the creative individual discourage productivity, whereas wars far away tend to encourage productivity (Simonton 1978). Simonton (1978) further emphasizes the need to look at conditions during childhood, adolescence, and early adulthood as a time during which “the creative genius is either made or broken” (p. 187), with its potential only fully actualised over the rest of the lifespan. He thus explores whether external events are more likely to influence people in the developmental period than in the productive period.

In our study, however, we are not interested in exploring the why question; our approach is more modest. That is, in the spirit of Gray’s (1966) 75 attempts to visualise creative cycles, we leverage a newly generated data set of over 5600 scientists across the last 200 years (1800–2000) to examine whether great minds actually tend to appear in batches. By providing insights on whether or not cycles actually exist, we contribute to the literature on the heated debate over nature versus nurture (Simonton 2013). If nature is the dominant force, the number of geniuses throughout history should be random; if nurture dominates, however, it should produce clear trends or cycles. Given prior emphasis on the fact that a quarter to a third of the variance in acquisition (performance) can be attributed to genetic factors (Simonton 2014a), non- randomness would suggest the need to give more thought to policy strategies

75 See also Kroeber (1944) and Sorokin (1957), as well as Turchin’s examinations of historical cycles (Turchin 2003; Turchin and Nefedov 2009).

207 that cultivate future geniuses. Nevertheless, exploring the genetic and environmental antecedents of genius and their potential interaction effects is methodologically challenging. On the other hand, Simonton (2014b, p. 614) asks the important question of what would happen were geniuses to vanish off the planet in the future. He also suggests an answer: “even when a civilization seems to enter a dark age, some other civilization seems always available to pick up where the dying civilization left off (p. 614). If we observe cycles rather than a downward trend in the number of geniuses, we can be more confident that the genius phenomenon will not cease to exist.

Measuring trends in great minds

Until now, any examination of genius trends or cycles has struggled to find solid measurements of creativity or eminence. For example, Simonton (2013, 2014a) emphasizes that geniuses offer original, useful and surprising ideas, while Gray (1966) is influenced by the opinions of Koeber, author of the Configurations of Culture Growth and pioneer of the systematic study of creativity in various civilizations: “I don’t think we can really measure creativity objectively. We can more or less measure opinions of it – the ratings are subjective, but their number or strength is more or less measurable. We’re then dealing with ‘reflections’ of the phenomena themselves; but that’s something. It is empirical and naturalistic” (p. 1391). Gray thus derives his proxies through ratings, differentiating between major, intermediate, and minor criteria, determined based on the following sets of questions:

• Major criteria “Has a man’s creation continued to be appreciated long after his era?” “Did his work reveal universal qualities of humanism?” “Did he rise above the limitations of his era?” “How influential was he on contemporary and subsequent creators?” • Intermediate criteria “How original was he?” “How versatile and many-sided was he?” “How prolific and sustained was his productivity?” “How great was his competence in the techniques of his art?” “In addition to form and beauty, did his work show social consciousness?” • Minor criteria “Was his work admired beyond his own country?” “Did he communicate, so that his work was contemporaneously popular?” (pp. 1391–1392)

208

The above questions clearly reflect the challenges that Gray faced in guaranteeing consistency throughout the measurement process. Recognizing the substantial impact that subjectivity has on ratings, he admits to certain limitations: “my ratings for creators of the period 1910–1935 must be regarded most dubiously… Contemporaneous evaluation of creators in all ages has been notoriously untrustworthy (remember that in their day people believed Cowley the peer of Milton, or Canova of Phidias)” (p. 1391). Scholars have thus tended to work with weighted measures that take into account the number of citations in archival sources (see, e.g., Simonton 1975a, b). A key issue in using such archival material, however, is finding an objective measure of which sources to consider and which to ignore. Simonton (1975b), for instance, relies heavily on general literary histories or anthologies, a shortcoming that we attempt to reduce by taking advantage of recent data made available through Google Books. These data allow us to provide new insights based on archival material that goes beyond anything implemented so far.

In general, it remains unclear whether or how creative cycles evolve over time. Scientists may be influenced not only by contemporaneous relationships but also by paragons active many years earlier. Gaps also exist between mentors and students or disciples (a 20-year age gap, according to Simonton (1984)), making a scientist’s social network highly complex and driven by factors within and between generations. This latter raises questions about the extent to which scientists are influenced by the current network, especially given that the half-life of knowledge and ideas is shorter in science than in art. The acknowledgement sections of major scientific books, for example, give the impression that some scientists are able to achieve a longer lasting

76 impression than suggested by the citation pattern of their articles. As Isaac Newton so aptly put it, “If I have seen further it is by standing on the shoulders of Giants”.

76 One analysis of a top economics journal demonstrates that the citation peak of articles is reached after 5 years (Johnston et al. 2013). McDowell (1982), in a discussion of the rate of knowledge obsolescence or depreciation reflected in the age profile of cited works in different disciplines, finds the largest decay rate in physics and chemistry (an annual average of 18.30), followed by economics (13.18), sociology and psychology (10.82), biology (8.68), history (3.85), and English (2.67).

209

Simonton (1984) also stresses that “[a]rtists, perhaps more than those in any other creative endeavour, are highly dependent on social networks for aesthetic success” (p. 1274). Nevertheless, he further observes that successful artists are strongly influenced by other artists who lived a long time before. Thus, preceding generations have a strong impact on contemporaries. For instance, in generations with three great minds, the artist paragon age gap in relation to artistic eminence is an inverted U-shape with the optimal eminence value at a gap of more than 200 years. On the other hand, people may “bootstrap” each other to higher achievements by influencing each other’s work. Hence, Yukawa (1978) asserts that scholars have a great effect on each other (as observed in school) by stimulating each other’s intellectual processes and providing an incentive to work hard and excel, both of which engender a proliferation of great minds. It is therefore a mixture of role- modelling, cohort effects, and external influences that shapes the structure of the eminence curve, making its form throughout history difficult to predict. Moreover, as James (1880) points out, because “community is a living thing,” social circumstances can make a genius incompatible with his surroundings; especially when “some previous genius of a different strain has warped the community away from the sphere of his possible effectiveness… Each bifurcation cuts off certain sides of the field altogether, and limits the future possible angles of deflection” (p. 447). It should also be noted that Gray (1966), in his analysis of Western civilization, disregards scientists and philologists, arguing that “in Western civilization, science was allied with technology, which underwent a linear rather than epicyclical evolution” (p. 1390). Our study, however, will show this assumption to be incorrect.

Materials and methods

To examine whether great minds or geniuses appear in cycles, we 77 employ the data from Bohannon (2011), which examines the literal fame of 5631 scientists born between 1800 and 1969. These data, drawn from Google Books, a digitised text corpus containing 4% of all books in print (Michel et al. 2011), show the number of times each scientist’s full name was mentioned

77 Data available from http://fame.gonzolabs.org/datasets (Adrian Veres is a co- contributor to the data set).

210 in English-language books published between 1800 and 2000. As a proxy for eminence, Bohannon (2011) standardises the frequency of name appearance by normalising it to the relative frequency of the name Charles Darwin. One Darwin unit thus represents the average annual frequency of a scientist’s full name relative to that of Darwin’s name. Because few scientists are mentioned as frequently as Darwin, however, with 82.5% of the 5630 scientists (excluding Darwin himself) having less than 0.01 Darwins of fame, Bohannon (2011) employs the milli-Darwin (one-thousandth of a Darwin, 78 mD). Based on this measure, Albert Einstein, with an mD of 878.2, has 87.82% (878.2/1000) of the mentions that Darwin has (which, by definition, is 1000 mD). Figure 9.1 shows the mD distribution for all 5,631 scientists and each of the five fields in which the scientists are working: chemistry, physics, 79 biology, mathematics, and social sciences. Interestingly, for all fields, we observe an exponential distribution: a Kolmogorov–Smirnov equality-of- distributions test for mD across all fields produces significant distributional differences across all fields except physics and chemistry (Table 9.1). Overall, the data set contains more scientists born in the twentieth century, particularly for physics, mathematics, and social science (see Figure 9.2). This proxy of eminence can also be seen as a proxy of creativity, measuring the extent to which scientists are valued during their lifetime and by succeeding generations. Hence, in the following analysis, we provide three statistical measures: mean, median, and total, based on the mD measure by each year of birth to assess whether certain periods are in fact characterised by higher levels of creativity.

78 See http://fame.gonzolabs.org/release-notes for the details on scientist selection and construction of the mD measure. 79 Five point 3% of scientists are classified into two or more fields.

211

Figure 9.1 Distribution of eminence across fields

212

Table 9.1 Kolmogorov–Smirnov equality-of-distributions test for mD across fields

Field 1 Field 2 p value Chemistry vs. Physics 0.626 Chemistry vs. Biology 0.017** Chemistry vs. Social Sciences 0.000*** Chemistry vs. Maths 0.023** Physics vs. Biology 0.032** Physics vs. Social Sciences 0.000*** Physics vs. Maths 0.000*** Biology vs. Social Sciences 0.000*** Biology vs. Maths 0.000*** Social Sciences vs. Maths 0.000*** Note: *, **, *** represent statistical significance at the 10%, 5% and 1% levels, respectively.

Figure 9.2 Number of scientists born each year

213

Results

In the Figure 9.3 depiction of our results, circles represent the mD (measured in natural log) generated by scientists, with only the names of the 50 highest ranked scientists shown. The blue line indicates a 10-year moving average of the total mD for scientists born in the same year (grey line = yearly value) using the elements of the tenth row of Pascal’s triangle (divided by the 80 horizontal sum of 512) as weight. We highlight the top 5 decades by designating the highest mean in blue and the total sum of mD in red, while shading overlapping periods in purple. The top 5 decades based on the mean mD are ranked in the following order: 1850–1859, 1804–1813, 1870–1879, 1827–1836, and 1895–1904; the top 5 decades based on the total sum of mD are ranked as follows: 1850–1859, 1900–1909, 1919–1928, 1878–1887, and 1936–1945.

For the first 100 years, a positive trend emerges that could support Gray’s (1966) claims of a linear relation with technological advances. However, contrary to Gray’s expectations, substantial cyclical fluctuation is evident around this trend. Specifically, as Figure 9.3 shows, many up and downs surround the following local peaks: 1812, 1822, 1832, 1843, 1851, 1858, 1872, 1879, 1884, 1896, 1902, 1916, 1922, 1927, 1933, 1935, and 1941. The average cycle is around 8.06 years with a standard deviation of 3.325 years. When looking at average values for 10-year intervals, we find that the most distinguished intellectuals, including Dewey, Freud, and Planck, were born during the 1850–1859 period. Interestingly, both Freud and Planck expressed their joy at the initial lack of attention among their colleagues, which allowed them to work in intellectual isolation, carefully developing the methods adopted or the knowledge created without fearing interference or competition (Merton 1973). The next most prolific period is 1804–1813, during which Darwin was born, followed by the 1870s, with the key examples of Einstein and Russell, who joined forces in the famous Einstein–Russell Manifesto (Schweber 2008). If we look at the sum of mD generated by different age cohorts, the 1850s account for 12.01% of all mD produced, followed by 1900–1909 (11.94%) when Oppenheimer, Pauling, and

80 For example, the 10-year moving average mD at time t equals 1 9 126 9 1 t * +t * +…+t* +…+t * +t * . -5 512 -4 512 512 +3 512 +4 512

214

Heisenberg were born, and 1919–1928 (9.58%), the period that produced Chomsky. The 1900–1909 cohort with Pauling and Heisenberg in the centre is linked to the birth of quantum chemistry, which arrived with the formulation of quantum mechanics in an era that revolutionised the physical sciences. All three scientists witnessed and were responsible for the overthrow of the classical concepts of space–time and of determinism in the description of atomic phenomena (Schweber 2003, p. 381). The period around the 1900s (1897–1906) is the most successful based on the percentage of scientists within the same age cohort who rank above the overall median mD. Nevertheless, the radical decay observed for the cohort born after 1940 should be treated with caution; giving them more time to “catch up” may lead to a change in the shape of the curve. Comparing our observations with those of Gray (1966), although he identifies the largest proportion of creative people (weighted by relative importance) in the 1890–1910 period (while expressing concern that the more recent creators might be overrated), he pinpoints the next most productive periods as 1815–1850 and 1850–1870, which is very close to our own findings. Gray also reports substantially lower values for the 1790–1815, 1870–1890, and post 1910 periods, which is relatively consistent with our observations. The distributions for each field, shown in Figure 9.4 and Figure 9.5, indicate more movement overall within single fields, particularly for social sciences.

We also recognise, however, that peaks can be produced by random outputs. We therefore use the non-parametric Wald–Wolfowitz test for randomness to assess whether the observations are serially independent (i.e., observed eminence occurs randomly) by counting runs above and below a threshold. In other words, we test whether the mD sequence follows a random order (autocorrelation is absent) or is mutually dependent (autocorrelation is present). Based on the assumption that eminence is positively serially correlated and tends to remain above or below its median or mean for several observations in a row (meaning fewer runs), we use the median and mean as our two thresholds. If the observations above the median/mean tend to be followed by observations below the median/mean, there will be relatively fewer runs. We first perform the runs test on the average and total mD generated by scientists born in the same year and the time series of the number of scientists born in the same year by counting the number of runs above or below the threshold. As the figures show, the time series plots of the average

215 mD, total mD, and number of scientists show steadily increasing values over time. Thus, we detrend the data to specify the thresholds. To obtain the detrended time series, we use the residuals from the regressions of the actual times series with mD as the dependent variable and the time variable (year of birth) as the independent variable. Observations equal to the threshold are treated as if they were below the threshold. These results remain robust to such changes as omission from the calculation or random assignment to groups above or below threshold.

216

Figure 9.3 Timeline of great minds born 1800–1969

217

We also note, however, that the time series seem to be downward biased towards the end of the time period (from 1945 onwards); we therefore truncate it by excluding scientists born after 1945. For the mD series, we use three factors—mean, sum, and count—and report results with and without smoothed values (see Table 9.2). Although those with smoothed values are all statistically significant, this finding should be treated with caution because smoothing reduces the number of runs (i.e., augments stability), which increases the statistical significance. For both thresholds, the median and the mean, the results without smoothing indicate a tendency not to reject the hypothesis of randomness. Thus, the results at the descriptive level should also be interpreted with care.

In a next step, we explore different percentiles to better understand whether non-randomness is more likely to occur among the greatest minds (Table 9.3). This trend is indeed observable, particularly for the top percentile group. To determine the yearly ratio of the number of scientists with an mD in the top X% among all scientists, we conduct a runs test on seven time series derived by assigning seven different values of X: 5, 10, 25, 50, 75, 90, and 95%. To illustrate, assuming 100 scientists born in the same year, 10 of whom have an mD in the top 5 percentile of all other scientists, then the ratio in that year for the top 5% is 10/100 = 0.1. Again, we perform the runs test using the median and mean of the time series as the thresholds but add an additional threshold at the ratio value of 0.5. Here, one run indicates a set of consecutive years in which the ratio of scientists in the top X% is above or below 0.5. In this case, by focusing on the top 25 scientists in each year, we demonstrate that the hypothesis of randomness can be rejected.

219

Table 9.2 Non-parametric test for non-randomness

Threshold mD series Number of Number below Number above Expected p value of z z statistic Number of Variance of the observations the threshold the threshold number of runs runs number of runs

Median Mean 145 72 73 73.50 0.36 -0.92 68 35.99 Sum 145 73 72 73.50 0.68 -0.42 71 35.99 Count 145 73 72 73.50 0.00 -3.08 55 35.99

Mean smoothed 141 71 70 71.50 0.00 -9.89 13 34.99 Sum smoothed 141 71 70 71.50 0.00 -9.21 17 34.99 Mean Mean 145 73 72 73.50 0.36 -0.92 68 35.99 Sum 145 75 70 73.41 0.46 -0.74 69 35.91 Count 145 79 66 72.92 0.00 -3.68 51 35.42 Mean smoothed 141 76 65 71.07 0.00 -9.88 13 34.57 Sum smoothed 141 78 63 70.70 0.00 -9.87 13 34.20

221

Table 9.3 Exploration of different mD percentiles

Threshold mD percentile Number of Number below Number above Expected p-value of z z statistic Number of Variance of the observations the threshold the threshold number of runs runs number of runs Median 5 145 73 72 73.50 0.04 -2.08 61 35.99 10 145 73 72 73.50 0.68 -0.42 71 35.99 25 145 78 67 73.08 0.25 1.16 80 35.58 50 145 77 68 73.22 0.90 0.13 74 35.72 75 145 73 72 73.50 0.93 -0.08 73 35.99 90 145 72 73 73.50 0.21 -1.25 66 35.99 95 145 73 72 73.50 0.36 -0.92 68 35.99 Mean 5 145 63 82 72.26 0.12 -1.57 63 34.76 10 145 62 83 71.98 0.73 0.34 74 34.49 25 145 62 83 71.98 0.39 0.85 77 34.49 50 145 68 77 73.22 0.90 0.13 74 35.72 75 145 73 72 73.50 0.93 -0.08 73 35.99 90 145 77 68 73.22 0.12 -1.54 64 35.72 95 145 79 66 72.92 0.51 -0.66 69 35.42 0.5 25 145 6 139 12.50 0.00 -3.82 9 0.84 50 145 77 68 73.22 0.90 0.13 74 35.72 75 145 144 1 2.99 0.00 -8.46 2 0.01

222

Conclusions

In answer to our initial research question, our results suggest that the clustering of great minds is probably not pure coincidence. Rather, our large data set of more than 5600 scientists over a 200 year period points to certain cycles of eminence. The periods 1850–1859 and 1897–1906 or 1900–1909, for example, were particularly prolific. The novelty in our analysis is that it takes advantage of Google’s massive archival effort Google Books, which allows us to work with data beyond those available to any previous analysis. It also focuses on the achievements of scientists, which most previous studies neglect based on the premise of an inherently linear (rather than cyclical) relation allied with technological advances. We, on the other hand, not only identify a positive trend in the 1800–1900 period but also observe cycles around this trend. Interestingly, these cycles observed for scientists are relatively similar to those observed for other eminent people, including cycles derived from data that rely heavily on subjective quality judgments. A non- parametric test of whether randomness can be rejected, however, indicates that for the raw (non-smoothed) data, non-randomness is less apparent. On the other hand, once we focus on the greatest minds, randomness is more likely to be rejected. In sum, these results imply that more research is needed to determine whether the greatest minds do indeed appear in batches or whether the apparent cycles are simply an artefact of random peaks.

A key limitation of our study is that we have worked with a dataset that only covers 4% of all books in print. Thus, the data analysed do not account for all the citations that the great minds have generated over the centuries. Future studies could show whether the 4% is a representative sample and whether the results obtained are robust. There are also, as discussed earlier, other aspects that should be considered. For example, few studies to date empirically examine trends in great minds from a historical perspective. Advances in technology will thus provide many areas for future research, and more detailed study of these larger trends should produce interesting observations. Advances in computing power, particularly, together with the growing capacity to work with Big Data, may prove to be to the social sciences what the microscope and telescope were to biology and astronomy, respectively (Aiden and Michel 2013). Because this area of

223 research is highly multidisciplinary, the resulting data sets should be capable of producing a richer, more realistic portrait of the social interactions of great minds and their consequences. In the future, for example, one could use the Google Books data to explore how sociocultural conditions and events such as zeitgeist, fractionalization, decentralization, economic conditions, war, civil disturbances, and political instability may affect the development of scientists’ creative potential.

Appendix

Figure 9.4 Eminence development by field based on mean mD

224

Figure 9.5 Eminence development by field based on sum mD

225

Chapter 10 Summary and Conclusions

This chapter summarises the findings and limitations from all eight studies of the thesis, and concludes by pointing out some directions for future research.

Summary of findings

In Chapter 2 and 3, I studied the award lifecycle of superstars in science using a novel collection of biographical data on major institutionalised awards acquired by 466 Nobel Prize laureates in Physics, Chemistry and Physiology or Medicine during 1901 to 2000. From the descriptive analysis presented in Chapter 2, I find a rapid increase in the number of major awards obtained by laureates during their scientific career, which reaches a pinnacle at the year of the Nobel Prize. However, once the Nobel Prize is conferred, a drastic drop in the number of awards received is observed, especially during the first 5 years after the Prize. This result is confirmed by multivariate analysis using the random effects negative binomial regression controlling for research field, gender, age, and scientific productivity over time. The key findings of this study suggest a possible Matthew effect in that award breeds award during eminent scientists’ careers when considering the rate of awards accumulation, but such effect, if any, disappears after the highest recognition is obtained, i.e. Nobel Prize. This is an empirical supplement to the current literature on cumulative advantage in science due to reception of awards, prize and honours, such as Chan et al. (2014) and Azoulay et al. (2014). The key results of the study in Chapter 3 extend on the previous chapter by examining how educational and methodological background are important factors in shaping the future recognition of prominent scientists. The comparison of the award life cycles between Nobel laureates allows us to hold subjects’ quality relatively constant (scientists judged to be the highest quality) while exploring factors that could explain the path of future recognitions. For example, compared with laureates who receive their education at other universities, earning a postgraduate degree from Cambridge, Harvard, or Columbia University generate 3.07 more

227

major awards throughout their academic career. Similarly, laureates with a theoretical orientation, as opposed to empirical, obtain 1.86 more awards. Completing one’s education in the US and UK is also beneficial for future recognition relative to completing education in other countries. These effects are quite large considering that Nobel laureates in the dataset generate, on average, 6.85 major awards throughout their career. Overall, the main contribution of these two studies is the original empirical evidence on the dynamic of individual recognitions through awards in the scientific community.

In Chapter 4, I investigated behavioural changes in research collaboration due to a sudden status shock, i.e. receiving the Nobel Prize. Specifically, I examine how receiving the Nobel Prize changes the recipients’ research collaboration patterns by analysing more than forty thousand coauthorship pairs of Nobel laureates from 1970 to 2000 and their collaborators. The results substantiate Zuckerman’s (1996) suggestion that the conferral of the Nobel Prize could have a decisive change in the collaboration formation and dissolution pattern. For instance, examining the number of new coauthors interacting with Nobel laureates on a yearly basis, I found a significant structural break at the Prize reception year, where the rate of change significantly changes from positive (before the Prize) to negative (after the Prize), even though the average number of new coauthors is higher in the post-Prize period. Moreover, the results demonstrated a high coauthor dropout before the Prize, suggesting that many of the coauthors who previously collaborated with the laureates cease interaction after they have received the Prize. I then compare the dropout rate (defined as percentage of current collaborators leaving the network) between collaborators who started interacting with laureates after the Prize to pre-award coauthors, and find that dropout rate of the latter is significantly lower. Further, in the multivariate analysis, I found that coauthors with longer cooperation history or higher pre- award collaboration intensity, as measured by the number of years between the Nobel Prize year and the first co-publication year and the number of pre- award publications, are less likely to leave the Nobel network before the Prize and have more post-award interactions with the laureate, thus suggesting relationship duration and intensity are important determinants in promoting collaboration loyalty and sustainability. In terms of collaborative research impact, while many previous studies have provided evidence on how research

228 teams are formed and their (citation) advantages over a solo-author product, yet, empirical results on the efficiency of academic team over time remains lacking in the literature. The study in Chapter 5 provided new insights into this research. Using the dataset from the previous chapter, I examined the intensity and citation performance of each collaboration pair. To isolate the effect of journal quality on citation performance, I used the residuals obtained from the first-stage regression (journal impact factor as dependent variable) as the control variable of journal visibility in the second stage of the two-stage least squares regression analysis. The results show that (on average) earlier collaborations attract significantly more citations than later publications from the same team when controlling for laureate and time fixed effect. Thus it is suggested that there is a diminishing return to impact, and perhaps also to creativity, from the intensity of collaboration between laureates and their coauthors. This finding suggests that research institutes should encourage the formation of new collaborations. For example, research agencies could develop platforms or portals that allow team building through proper matching based on researchers’ skillsets, interests, and project needs, which could lower the search cost of finding suitable co-authors thus promote research collaborations and team formation efficiency.

In Chapter 6, I explored the correlation between scholarly impact and the external influence of top economists. The external influence of scholars is proxied by the number of Google and Bing search results conducted. Using 723 of the top 1000 economists from the RePEc ranking as sample, I found a large discrepancy between the rankings based on external influence and internal scholarly impact, which echoes the findings of Aguinis et al. (2012) in management science. In addition, further examinations with various measures of publication and citation performance confirms a consistent positive but weak correlation between external influence and internal scholarly impact. Such results may call for a reconsideration on universities’ reward system on researcher performance to give more weights on activities that generate larger societal and social scholarly impact (such as industrial engagement or giving public lectures), if the university wish to increase its overall public engagement and impact.

In Chapter 7, I extend the results from the previous chapter to examine whether scholars can capitalise on their external and internal influence for those who participate in the speaking market. I analyse the

229

speaking fees charged of a sample of 580 speakers with different levels of academic involvement in different disciplines (full-time, part-time or non- academics). Controlling for speakers’ individual characteristics, our results indicate that scholars with more external influence outside of academia charge a higher (minimum) speaking fee. The results are consistent when different proxies for external prominence were used, such as number of non- .edu web pages, number of books listed on the Library of Congress, invitations to TED talks, receiving a book award, listing on The New York Times Best Sellers list. On the contrary, internal academic success does not explain the level of speaking fees once number of Google pages is controlled for, which provides a clear distinction between the capitalization of external and internal influence in the speaking market. To the best of my knowledge, this is the first research examining the determinants of speaking fees.

In Chapter 8, I assessed whether observable citation performance bias due to institutional (doctoral or work affiliation) prestige in the short run diminishes over time. In particular, based on the assumption that scientific articles published at the same time (same issue) in the same outlet (same journal) should have a similar impact (subsequent citation trajectory), I tested the citation differences between papers by authors with different educational and work affiliation background who were published in American Economic Review and Econometrica. Overall, I found evidence supporting a systematic positive citation bias to articles authored by economists from highly ranked universities. Such bias disappeared in the long run. This evidence challenges the validity of citation-based metrics in terms of research evaluation. An important implication for science policy is the question how to evaluate the citation performance of researchers based on this result. Specifically, universities or funding agencies may take into account such bias (citations due to institutional reputation as oppose to the quality of the research) when evaluating the potential candidates for job hiring, promotion and funding grants applications, or rely less on citation indices when making such decisions.

In the final study in Chapter 9, I provide an historical overview of the emergence of scientific great minds using a novel dataset with more than 5,600 eminent scientists over two centuries. Seeing ‘literal’ fame of scientists as a proxy measurement of creativity, I observed a positive increasing trend during the 19th century. Also, looking at the periods when scientific genius

230 was particularly abundant, there seem to be a cycle fluctuating around the positive trend. In an attempt to determine whether there is actually a cyclical pattern to the appearance of great scientific minds, I conducted a Wald– Wolfowitz runs test in order to check whether the observed eminence are serially independent. The results suggest that such a process is not likely to a coincidence.

Overall, the collection of works presented in this thesis has contributed empirically to the scientometrics literature in regards to recognition through awards and citations, status shock, collaboration behaviour and pattern, academic societal impact, citation bias in the long run, and the emergence of great minds.

Shortcomings

While some of the shortcomings have already been discussed in each individual chapter, I reiterate the key limitations and provide additional noteworthy shortcomings of the studies collected in this thesis.

Several studies are explorative in nature where analysis does not include a comparison group, which allows no inferences to be drawn about associations, causal or otherwise. Yet, as emphasized by distinguished Australian scholar Adrian Pagan, it is important to have both summative (e.g. extrapolate the facts) as well as interpretative (e.g. explain the facts using precise models) research. Therefore, I personally do not see this (descriptive approach) as a limitation, as I do not draw causal inference when it is not applicable. I see it rather as a springboard into more rigorous studies with appropriate control groups. Thus, it is important to understand that this thesis does not aim to be definitive, but rather, it seeks to serve as a primer for future research on this topic.

In addition, even though the studies in this thesis took advantage of using top scholars (Nobel laureates, highly ranked economists, research published in top economic journals and renowned scientific figures) as homogenous groups within a very controlled setting (i.e. academia), these subjects are highly successful and prolific and thus may not be representative of the broader academic population, especially given that success in the academic market is very skewed. Thus, one should exercise caution with the external validity of our results when attempting to draw implications.

231

Chapter 2 and 3: Recognition cycle with academic awards In the analysis of both studies, there is no weighting for the different major institutional awards and each award is assumed to be equal. Yet, the perceived value of each award is likely to be different. The question is then how to measure the relative importance of an award so that the award path of scientist can be better depicted. Perhaps one could employ a similar approach as Bohannon (2011) to use the Nobel Prize as a baseline and calculate the relative frequency of mentions among literature (books or scientific articles), social media or other channels (Google hits) as a proxy of the importance of the award, or at least how well-known is it. Such approach could incorporate more minor awards that were ignored in the study. Another possibility is to then see whether less popular awards tend to (quickly) follow the more important ones, such as a sudden increase in the number of honorary titles after receiving a major award. As discussed in the chapter, as one of the key aspects of awards is supply driven, the type of award (field/methodological/ location specific), availability and the award institution characteristics in our findings are not clear and could be controlled for in future studies. One could also conduct surveys among committee members of awards to better understand their decision process.

Chapter 4 and 5: Collaboration pattern and productivity Although I find a systematic and distinct structural break on research collaboration practice, the study does not inform whether such abrupt change was due to decision from the recipient or from the existing or future/potential collaborators or both. Such a question is inherently difficult to provide empirical evidence for, yet it is an important one to address. A related issue is the change of affiliation of the Nobel laureate after the Prize; this is also not controlled for.

Another shortcoming of the study in Chapter 4 (and 5) was that I did not examine collaborators’ individual characteristics, such as their (academic) age, gender and work affiliation/location, which may have significant effects on the collaboration pattern of Nobel laureates. For example, examining the individual characteristics of the coauthors could provide an idea of whether laureates had switched to a more supervisory role after receiving the Prize. In some cases, the average age of the coauthors dropped significantly, which could be due to supervising younger students, or it could be that coauthors stop collaborating at a certain point in time due to retirement, relocation of

232 work, or coauthor being deceased. A potential channel to obtain such data would be from the Doctoral dissertation database (such as ProQuest) and the CV of the researchers. Research specific attributes such as the size of the project, methodological approach, or research interest could also have large implications for the research team size, interaction and impact: thus controlling for them could provide more precise estimation results. Potential ways to obtain these factors are by checking with the research grant applications or conducting a content analysis on the research output (locating keywords, figure, tables or even mathematics used).

The proportion of solo-author papers throughout a Nobel laureates’ career relative to the number of collaborative works was also excluded from the study. One could examine whether there is also a structural break on the number of single author articles by a laureate before and after the Prize and how it correlates with the collaborative behaviour. The reason for such change, if it exists, should also be identified: whether it is due to change of research topic (exploring new fields, which could also have very different citation trajectory) or methodological approach (e.g. switch to providing theories) or other factors.

Since the time component is defined as the year of publication, the actual time when collaborations start is not known. Moreover, due to different time lags in the publication process, year of publication does not necessary equal the end of the collaboration. Therefore, there is a chance of both over- and under-estimation towards the length of collaboration history.

Chapter 6 and 7: External and internal influence One of the major limitations in the two studies on external influence is the inability to control for the dynamic relationship between external and internal impact, since the studies only provide a snapshot of both measures, and the model is static in nature. The endogenous link between the two measurements could be examined in future studies. Moreover, the measure of external impact is direct, i.e. it does not take into account the indirect external impact via internal influence where one’s influence on other scholars’ work, which itself would generate external influence. Thus, the findings could be biased since the samples’ research interests/topics/approach are not controlled for. For example, a theoretically driven study could provide important insights that enable other scholars’ work (internal), which has external impact

233

(indirect external impact of the theoretical study which is not captured). Yet, it is unclear when such a link would break. A remedy is validity check with alternative proxies. Nevertheless, a deeper exploration considering the citation/research network of scientists could provide a better insight.

It is possible that there might be self-selection bias in the sample used in Chapter 7, as academics who enter the speaking market might have quite distinct characteristics, relating to personality, language (and public speaking skills) or internal academic activity (research topics) and performance, compared to researchers who do not engage in public activities. Similar to the impact measures, the speaking fees captured are also static in nature. Furthermore, I do not know the speakers’ experience in the speaking market and whether the speaking fee demanded is the result of an adjustment process, even though speakers’ professional age and TED talk experience is controlled for. Additionally, the demand for speakers’ knowledge and expertise (or fame) might vary between the topics of interest and could have some implications on the study’s findings.

Chapter 8: Quality of economic research in the long run One of the limitations in these chapters is the reliance on a static university ranking to model perceived institutional prestige, which could be a dynamic process in nature. For instance, since I model citation count per year, the perceived affiliation prestige of authors back in 1960 could be different to todays’ view. Since the ranking was created recently, such a proxy might be less reliable with respect to the earlier years. Another major drawback is that the study did not control for the topic of each research article. Even though the sample journal Econometrica has a specific focus, different research topics could have large implications for the article’s citation trajectory. Controlling for the topic could improve the validity of the existing results. For example, future research could control for research topics using the JEL code of the article or conduct content analysis to find comparable articles (e.g. co-word analysis or keyword extraction on the abstract or the entire article). Another extension is to analyse the content of the citing articles and the respective authors.

Chapter 9: Great minds Since the data was drawn from only books published in the English language, the proxy of creativity may be subject to bias (towards Western

234 scientists). A visual plot by scientists’ country of origin or usual residence could provide a starting point. Future research could also overcome this by incorporating Google Books’ corpus of other languages. This could also enable a study to examine the scientific knowledge exchange or influence between cultures and countries.

Direction for future research

In the end, this thesis raises many questions and offers opportunities which future research in science knowledge and scientometrics can build on.

Apart from the suggestions outlined in the limitation sections, the most obvious recommendation for future research is to extend the subject of interest from the most prolific scientists to the rest of the labour force. Continuous efforts are needed to expand the area of scientometrics by increasing the accessibility to more comprehensive data on scientists to allow a complete mapping of scientific activities and thus progress. While many studies on scientometrics provide insightful results using static cross- sectional data (a snapshot in time), I call for further research into problems that are more dynamic in nature, by using longitudinal data and network structured data.

A simple extension to the study in Chapter 2 is to collect the list of scientists who also receive the major awards in our database along with their publication and citation data. One could then estimate the casual effect of obtaining the Nobel Prize on the award life cycle, using the synthetic control method for finding a comparable control group base on the age-performance and age-award data. Another important topic need better understand is the timing of award. For example, Merton (1973) stresses the importance of understanding the most effective timing of major awards to incentivise excellent research. Suboptimal timing (too early or too late) of award might bring dissatisfaction, frustration, and pressure to perform which could have long-term consequences. Drawing on award life cycle data on Nobel laureate and other scientists, one can assess the implications of how premature or delayed recognition impact scientists’ longevity, retirement choices, performance, and posthumous recognition (fame immortality). Furthermore, the first part of the result, which implies potential inequality allocations of awards encouraging a superstar market, may have some policy relevance for award organisations to enhance effectiveness of award structures. For

235

example, future research can examine topics such as the marginal benefits and costs of awards, the proper timing of awards (early versus late career awards) so that we have a better understanding how to supply awards so as to promote creativity, innovation, effectiveness, and motivation.

In order to gain a more complete picture on collaboration dynamics, I will expand the data by obtaining the collaboration network of the coauthors of Nobel laureates. One of the main benefits is to isolate the unobservable effect to the existing analysis such as controlling for the capacity of each co-author, as measured by the number of current co-author at a given time, productivity level before and after joining the Nobel Prize network. Furthermore, one could test whether having work with a super star (Nobel laureate) can increase the status, performance, network size and job mobility of the co-author. It can draw implication on whether coauthors were able to capitalise the work experience with a super star, even after leaving the star pool? Lastly, the dataset with 5,631 eminent scientists can be expanded by incorporating additional background information and productivity data for each scientists. A detailed mapping of these great minds would allow us to study positive and negative life shocks on scientists and how knowledge diffuse between disciplines.

236

Bibliography

Abramo, G., D’Angelo, C. A., & Murgia, G. (2013). Gender differences in research collaboration. Journal of Informetrics, 7, 811–822. Acemoglu, D. (1995). Reward structures and the allocation of talent. European Economic Review, 39(1), 17–33. Adams, C. W. (1946). The age at which scientists do their best work, Isis, 36, 166–69. Adams, J. D., Black, G. C., Clemmons, J. R., & Stephan, P. E. (2005). Scientific teams and institutional collaborations: Evidence from U.S. universities, 1981–1999. Research Policy, 34, 259–285. Aguinis, H., Suarez-Gonzales, I., Lannelongue, G., & Joo, H. (2012). Scholarly impact revisited. Academy of Management Perspectives, 26(2), 105–132. Aiden, E., & Michel, J.-B. (2013). Big data as a lens on human vulture. New York: Riverhead Books. Albert, R. S. (1975). Toward a behavioral definition of genius. American Psychologist, 30, 140–151. Allison, P. D. & Stewart, J. A. (1974). Productivity differences among scientists: Evidence for accumulative advantage, American Sociological Review, 39, 596–06. Allison, P. D. (1980). Inequality and scientific productivity. Social Studies of Science, 10, 163–179. Allison, P. D., Long, J. S. & Krauze, T. K. (1982). Cumulative advantage and inequality in science, American Sociological Review, 47, 615–25. Alston, R. M., Kearl J. R. & Vaughan M. B. (1992), Is there a consensus among economists in the 1990s?. American Economic Review 82, 203– 9. Amir, R., & Knauff, M. (2008). Ranking economics departments worldwide on the basis of PhD placement. Review of Economics and Statistics, 90, 185–190. Arrow, K. J., Bernheim B. D., Feldstein M. S., McFadden D. L., Poterba J. M. & Solow R. M. (2011), 100 years of the American Economic Review: The top 20 articles, American Economic Review 101, 1–8.

237

Aubrey, J. (1898/2007). Lives of eminent men. London: Hesperus Classics. Aversa, E. S. (1985). Citation patterns of highly cited papers and their relationship to literature aging: A study of the working literature. Scientometrics, 7(3–6), 383–389. Axarloglou, K., & Theoharakis, V. (2003). Diversity in economics: An analysis of journal quality perceptions. Journal of the European Economic Association, 1(6) 1402–1423. Azoulay, P., Stuart, T., & Wang, Y. (2014). Matthew: Effect or fable? Management Science, 60(1), 92–109. Baffes, J., & Vamvikidis, A. (2011). Are you too young for the Nobel Prize? Research Policy, 40, 1345–1353. Baldi, S. (1998). Normative versus social constructivist processes in the allocation of citations: A network-analytic model. American Sociological Review, 63: 829–846. Barabási, A. L. (2003), Linked. How Everything is Connected to Everything Else and What It Means for Business, Science, and Everyday Life, Plume Books, New York. Barabási, A.-L. & Albert R. (1999), Emergence of scaling in random networks, Science 286, 509–12. Barbezat, D. A. (1987). Salary differentials or sex discrimination? Evidence from the academic labor market. Population Research and Policy Review, 6(1), 69–84. Barbezat, D. A. (1991). Updating estimates of male-female salary differentials in the academic labor market. Economics Letters, 36(2), 191–195. Bar-Ilan, J. (2010). Citations to the “introduction to informetrics” indexed by WOS. Scopus and Google Scholar. Scientometrics, 82(3), 495–506. Barjak, F., Li, X., & Thelwall, M. (2007). Which factors explain the web impact of scientists’ personal homepages? Journal of the American Society for Information Science and Technology, 58(2), 200–211. Bauer, K., & Bakkalbasi, N. (2005). An examination of citation counts in a new scholarly communication environment. D-Lib Magazine, 11(9), www.dlib.org/back2005.html. Accessed 26 May 2014. Baumol, W. J. (2000). What Marshall didn’t know: On the twentieth century’s contributions to economics. Quarterly Journal of Economics 115, 1–44.

238

Beaver, D. D. (2001). Reflections on scientific collaboration, (and its study): Past, present, and future. Scientometrics, 52, 365–377. Becker, M. (1995). Nineteenth-century foundations of creativity research. Creativity Research Journal, 8(3), 219–229. Bernal, J. D. (1939). The social function of science. New York: Macmillan. Biswal, A. K. (2013). An absolute index (Ab-index) to measure a researcher’s useful contributions and productivity. PLoS ONE, 8(12), e84334. Bjork. S., Offer, A., & Söderberg, G. (2014). Time series citation data: the Nobel Prize in economics. Scientometrics, 98(1), 185–196. Blaug, M. (1997). Ugly Currents in Modern Economics. Policy Options, Options Politiques 18,3–8. Bohannon, J. (2011). The science hall of fame. Science, 331(6014), 143. Bollen, J., Van de Sompel, H., Hagberg, A., & Chute, R. (2009). A principal component analysis of 39 scientific impact measures. PLoS One, 4(6), e6022. Bonitz, M., Bruckner, E. & Scharnhorst, A. (1997). Characteristics and Impact of the Matthew Effect for Countries, Scientometrics, 40: 407– 22. Borjas, G. J., & Doran, K. B. (2015). Prizes and productivity how winning the fields medal affects scientific output. Journal of Human Resources, 50(3), 728-758. Börner, K., Contractor, N., Falk-Krzesinski, H. J., Fiore, S. M., Hall, K. L., Keyton, J., et al. (2010). A multilevel systems perspective for the science of team science. Science Translational Medicine, 2(49), 49cm24. Bornmann, L. & Daniel, H. D. (2008). What do citation counts measure? A review of studies on citing behaviour, Journal of Documentation, 64, 45–80. Bornmann, L., & Daniel, H.-D. (2007). What do we know about the h index? Journal of the American Society for Information Science and Technology, 58(9), 1381–1385. Bothner, M. S., Haynes, R., Lee, W. & Smith, E. B. (2010). When do Matthew Effects Occur?, Journal of Mathematical Sociology, 34: 80–114. Bothner, M. S., Podolny, J. M. & Smith, E. B. (2011). Organizing Contests for Status: The Matthew Effect vs. the Mark Effect’, Management Science, 57: 439–57.

239

Bourdieu, P. (1989). Social space and symbolic power. Sociological Theory, 71(1), 14–25. Bozeman, B., & Corley, E. (2004). Scientists’ collaboration strategies: Implications for scientific and technical human capital. Research Policy, 33, 599–616. Bratsberg, B., Ragan, J. F, Jr, & Warren, J. T. (2003). Negative returns to seniority: New evidence in academic markets. Industrial and Labor Relations Review, 56(2), 306–323. Brittan, S. (1973), Is There an Economic Consensus? An Attitude Survey, Macmillan, London. Buchanan, J. M. (2000). Saving the Soul of Classical Liberalism. Wall Street Journal, R29. Cameron, A. C., & Trivedi, P. K. (2009). Microeconometrics using Stata. College Station: A Stata Press Publication. Cameron, B. D. (2005). Trends in the usage of ISI bibliometric data: Uses, abuses, and implications. Libraries and the Academy, 5(1), 105–125. Cattell, J. M., & Brimhall, D. R. (1921). American men of science: A biographical directory. Garrison, NY: The Science Press. Certo, S. T., Sirmon, D. G., & Brymer, R. A. (2010). Competition and scholarly productivity in management: Investigating changes in scholarship from 1988 to 2008. Academy of Management Learning and Education, 8(4), 591–606. Chan, H. F. & Torgler B. (2012), Econometric Fellows and Nobel Laureates in Economics, Economics Bulletin 32, 3365–77. Chan, H. F., & Torgler, B. (2013). Time-lapsed awards for excellence. Nature, 300, 29. Chan, H. F., & Torgler, B. (2015). The implications of educational and methodological background for the career success of Nobel laureates: An investigation of major awards. Scientometrics, 102, 847–863. Chan, H. F., Frey, B. S., Gallus, J., & Torgler, B. (2014). Academic honors and performance. Labour Economics, 31, 188–204. Chan, H. F., Frey, B. S., Gallus, J., Schaffner, M., Torgler, B., & Whyte, S. (2016). External influence as an indicator of scholarly importance. CESifo Economic Studies, 62(1), pp. 170-195 Chan, H. F., Frey, B. S., Gallus, J., Schaffner, M., Torgler, B., & Whyte, S. (2014). Do the best scholars attract the highest speaking fees? An

240

exploration of internal and external influence. Scientometrics, 101(1), 793–817. Chan, H. F., Gleeson, L., & Torgler, B. (2014). Awards before and after the Nobel Prize: A Matthew effect and/or a ticket to one’s own funeral. Research Evaluation, 23, 210–220. Chan, H. F., Önder, A. S., & Torgler, B. (2015). Do Nobel laureates change their patterns of collaboration following the prize reception? Scientometrics, 105(3), 2215–2235. Chan, H. F., Önder, A. S., & Torgler, B. (2016). The first cut is the deepest: Repeated interactions of coauthorship and academic productivity in Nobel laureate teams. Scientometrics, 106(2), 509–524. Chubin, D. E., Porter, A. L., & Rossini, F. A. (1984). “Citation classics” analysis: An approach to characterizing interdisciplinary research. Journal of the Association for Information Science and Technology, 35, 360–368. Clower, R. W. (1993). The State of Economics: Hopeless but not serious? in D. Colander and A. W. Coats, eds, The Spread of Economic Ideas. Cambridge University Press, Cambridge, pp. 23–30. Coase, R. H. (1994), Essays on Economics and Economists, Chicago University Press, Chicago and London. Cole S. (1970). Professional standing and the reception of scientific discoveries. American Journal of Sociology, 76, 286–306. Cole, J. & Cole S. (1971). Measuring the Quality of Sociological Research: Problems in the use of the Science Citation Index. American Sociologist 6, 23–9. Cole, J. & Singer, B. (1991). A theory of limited differences: Explaining the productivity puzzle. In J. C. Harriet Zuckerman & J. Bauer (eds.), The Outer Circle: Women in the Scientific Community. New York: W.W. Norton and Company, 277–340. Cole, S. (1970). Professional Standing and the Reception of Scientific Discoveries. American Journal of Sociology, 76: 286–306. Cole, S. (1979). Age and Scientific Performance. American Journal of Sociology, 84: 958–77. comScore (2014). August 2014 U.S. Search Engine Rankings. September 17, http://www.comscore.com/Insights/Market-Rankings/comScore- Releases-August-2014-US-Search-Engine-Rankings (last accessed 18 December 2014).

241

Coupe, T., Ginsburgh V. & Noury A. (2010). Are Leading Papers of Better Quality? Evidence from a natural experiment. Oxford Economic Papers 62, 1–11. Cox, R. A. K., & Chung, K. H. (1991). Patterns of research output and author concentration in the economics literature. Review of Economics and Statistics, 73(4), 740–747. Crichton, M. (1999). Ritual abuse, hot air, and missed opportunities. Science, 283(5407), 1461–1463. Cronin, B. (2001). Hyperauthorship: A postmodern perversion or evidence of a structural shift in scholarly communication practices? Journal of the American Society for Information Science and Technology, 52(7), 558– 569. Dasgupta, P. (1998), Modern Economics and Its Critics, mimeo, University of Cambridge, Cambridge. Dasgupta, P., & David, P. A. (1994). Toward a new economics of science. Research Policy, 23(5), 487–521. Diamond, A. M, Jr. (1986). What is a citation worth. Journal of Human Resources, 21(2), 200–215. Dietz, J. S., & Bozeman, B. (2005). Academic careers, patents, and productivity: Industry experience as scientific and technical human capital. Research Policy, 34, 349–367. Dixit, A. (1994). My system of work (Not!). American Economist, 38(1), 10– 16. Ductor, L. (2014), Does co-authorship lead to higher academic productivity? Oxford Bulletin of Economics and Statistics. Duncan, K. C., Krall, L., Maxcy, J. G., & Prus, M. J. (2004). Faculty productivity, seniority, and salary compression. Eastern Economic Journal, 30(2), 293–310. Dunwoody, S., & Ryan, M. (1985). Scientific barriers to the popularization of science in the mass media. Journal of Communication, 35(1), 26– 42. Egghe, L. (2006). Theory and practice of the g-index. Scientometrics, 69(1), 131–152. Einstein, A., & Infeld, L. (1938). The Evolution of Physics. Cambridge: Cambridge University Press. Engemann, K., & Wall, H. J. (2009). A journal ranking for the ambitious economist. Federal Reserve Bank of St. Louis Review, 91, 127–39.

242

Eom, Y.-H., Fortunato, S. (2011). Characterizing and modeling citation dynamics. PLoS ONE, 6(9), e24926. Erren, T. C. (2008). Hamming’s “open doors” and group creativity as keys to scientific excellence: The example of Cambridge. Medical Hypotheses, 70(3), 473–477. Evans, K. (2009). The New Stars of the Blogosphere. Wall Street Journal, D1. Finkenstaedt, T. (1990). Measuring research performance in the humanities. Scientometrics, 19(5–6), 409–417. Fischhoff, B., & Scheufele, D. A. (2013). The science of science communication. Proceedings of the National Academy of Sciences, 110(Supplement 3), 14031–14032. Fox, M. F. (1983). Publication productivity among scientists: A critical review. Social Studies of Science, 13(2), 285–305. Franck, G. (1999). Scientific communication: A vanity fair? Science, 286(5437), 53–55. Frank, R. H. & Cook, P. J. (1996) The Winner-Take-All Society. New York: Free Press. Frey, B. S. & Gallus, J. (2014) Awards are a special kind of signal, CREMA Working Paper Series 2014-04. Center for Research in Economics, Management and the Arts (CREMA). Frey, B. S. & Neckermann, S. (2009). Awards: A view from economics. In: Brennan, G and Eusepi, G (eds) The Economics of Ethics, pp. 73–88. Cheltenham, UK: Edward Elgar. Frey, B. S. & Osterloh, M. (2010). Motivate people with prizes. Nature, 465: 871. Frey, B. S. (2006). Giving and receiving awards. Perspectives on Psychological Science, 1: 377–88. Frey, B. S. (2006). How influential is economics? De Economist, 154(2), 295–311. Frey, B. S. (2007). Awards as compensation. European Management Review, 4(1), 6–14. Frey, B. S., Pommerehne W. W., Schneider F. & Gilbert G. (1984). Consensus and Dissension among Economists: An empirical inquiry. American Economic Review 74, 986–94. Friedman, M. (1972). Have monetary policies failed?. American Economic Review 62, 17–18.

243

Frodeman, R., & Holbrook, J. B. (2007). Science’s social effects. Issues in science and technology, 23(3), 28–30. Galbraith, J. K. (1975). Letter. The Times, 16 July, p. 12. Galton, F. (1874). English men of science: Their nature and nurture. London: MacMillan. Garfield, E. (1989). More delayed recognition Part 1. Examples from the genetics of color blindness, the entropy of short-term memory, phosphoinositides, and polymer rheology. Current Contents, 38, 3–8. Garfield, E. (1990). More delayed recognition Part 2. From inhibin to scanning electron microscopy. Current Contents, 9, 3–9. Gergen, K. J. (1999). An invitation to social construction. Thousand Oaks, CA: Sage. Gilbert, G. N. (1977). Referencing as persuasion. Social Studies in Science, 7, 113–122. Gingras, Y., & Wallace, M. L. (2010). Why it has become more difficult to predict Nobel Prize winners: A bibliometric analysis of nominees and winners of the chemistry and physics prizes (1901–2007). Scientometrics, 82, 401–412. Ginsburgh, V. A. (2003). Awards, success and aesthetic quality in the arts. Journal of Economic Perspectives, 17(2), 99-11. Glänzel, W. & Schubert, A. (1985). Price Distribution. An Exact Formulation of Price’s “Square Root Law”. Scientometrics, 7: 211–19. Glänzel, W. (2006). On the opportunities and limitations of the H-index. Science Focus, 1(1), 10–11. Glänzel, W., & Schoepflin, U. (1994). Little scientometrics, big scientometrics … And beyond? Scientometrics, 30(2–3), 375–384. Glänzel, W., Schlemmer, B., Thijs, B. (2003) Better late than never? On the chance to become highly cited only beyond the standard bibliometric time horizon. Scientometrics, 58(3), 571–586. Golder, S. A., & Macy, M. W. (2011). Diurnal and seasonal mood vary with work, sleep, and daylength across diverse cultures. Science, 333(6051), 1878–1881. Goldstone, J. A. (1979). A deductive explanation of the Matthew effect in science. Social Studies of Science, 9: 385–91. Gomez-Mejia, L. R., & Balkin, D. B. (1992). Determinants of faculty pay: An agency theory perspective. Academy of Management Journal, 35(5), 921–955.

244

Gowan, J. C., & Olson, M. (1979). The society which maximizes creativity. Journal of Creative Behavior, 13(3), 194–210. Gray, C. E. (1966). A measurement of creativity in western civilization. American Anthropologist, 68(6), 1384–1417. Greene, M. (2007). The demise of the lone author. Nature, 450, 1165. Gross, P. L. K., & Gross, E. M. (1927). College libraries and chemical education. Science, 66(1713), 385–389. Guimera, R., Uzzi, B., Spiro, J., & Amaral, L. A. N. (2005). Team assembly mechanisms determine collaboration network structure and team performance. Science, 308, 697–702. Gupta, H. M., Campanha, J. R., & Pesce, R. A. G. (2005). Power-law distributions for the citation index of scientific publications and scientists. Brazilian Journal of Physics, 35(4), 981–986. Hall, B. H., Jaffe A. B. & Trajtenberg M. (2001). The NBER Patent Citations Data File: Lessons, insights and methodological tools. NBER Working Paper No. 8498, National Bureau of Economic Research, Cambridge, MA. Ham, J. C., & Weinberg, B. A. (2011). Geography and innovation: Evidence from Nobel laureates. Unpublished manuscript. Hamermesh, D. S. & Schmidt P. J. (2003). The determinants of Econometric Society Fellows elections. Econometrica 71, 399–407. Hamermesh, D. S., Johnson, G. E., & Weisbrod, B. A. (1982). Scholarship, citations and salaries: Economic rewards in economics. Southern Economic Journal, 49(2), 472–481. Hansen, W. L., Weisbrod, B. A., & Strauss, R. P. (1978). Modeling the earnings and research productivity of academic economists. Journal of Political Economy, 86(4), 729–741. Harzing, A.-W. (2010). The publish or perish book: Your guide to effective and responsible citation analysis. Melbourne: Tarma Software Research Pvt Ltd. Harzing, A.-W., Alakangas, S., & Adams, D. (2014). hIa: An individual annual h-index to accommodate disciplinary and career length differences. Scientometrics, 99(3), 811–821. Hayek, F. A. (1991). On being an economist. in W. W. Bartley and S. Kresge, eds, The Trend of Economic History, University of Chicago Press, Chicago, pp. 35–48.

245

Heinze, T., & Kuhlmann, S. (2008). Across institutional boundaries? Research collaboration in German public sector nanoscience. Research Policy, 37, 888–899. Henrekson, M., & Waldenström, D. (2011). How should research performance be measured? A study of Swedish economists. The Manchester School, 76(6), 1139–1156. Hirsch, B. T., Austin, R., Brooks, J., & Moore, J. B. (1984). Economics departmental rankings: Comments. American Economic Review, 74(4), 822–826. Hirsch, J. E. (2005). An index to quantify an individual’s scientific research output. Proceedings of the National Academy of Sciences, 102(46), 16569–16572. Hjørland, B. (2013). Citation analysis: A social and dynamic approach to knowledge organization, Information Processing & Management, 49(6), 1313–1325. Hogan, T. D. (1986). The publishing performance of U.S. Ph.D. programs in economics during the 1970s. Journal of Human Resources, 21(2), 216– 229. Holbrook, J. B., Barr, K. R., & Brown, K. W. (2013). We need negative metrics too. Nature, 497, 439. Hollis, A. (2001). Co-authorship and the output of academic economists. Labour Economics, 8(4), 503–530. Hosp, G. & Schweinsberg K. (2006). Für eine Handvoll Euros: Der Markt für Vorträge zur Bewertung des Einflusses der Volkswirtschaftslehre. Perspektiven der Wirtschaftspolitik 7, 459–69. Hosp, G., & Schweinsberg, K. (2006). Für eine handvoll Euros: Der Markt für Vorträge zur Bewertung des Einflusses der Volkswirtschaftslehre. Perspektiven der Wirtschaftspolitik, 7(4), 459–469. Howe, M. J. A. (1999). Genius explained. Cambridge: Cambridge University Press. Hsu, J.-W., Huang, D.-W. (2011). Dynamics of citation distribution. Computer Physics Communications, 182(1), 185–187. Huang, W. (2015). Do ABCs get more citations than xyzs? Economic Inquiry, 53, 773–789. Hudson, J. (2007). Be known by the company you keep: Citations—quality or chance? Scientometrics, 71, 231–238.

246

Hunt, J. G. & Blair, J. D. (1987). Content, process, and the Matthew effect among management academics. Journal of Management, 13: 191–10. James, W. (1880). Great men, great thoughts and the environment. Atlantic Monthly, 46(276), 441–459. Jensen, P., Rouquier, J.-B., Kreimer, P., & Croissant, Y. (2008). Scientists who engage with society perform better academically. Science and Public Policy, 35(7), 527–541. John, P. (2012). Debate: How relevant is political studies in the UK? Political Insights, 3(2), 16–18. Katz, D. A. (1973). Faculty salaries, promotions, and productivity at a large university. American Economic Review, 63(3), 469–477. Johnston, D. W., Piatti, M., & Torgler, B. (2013). Citation success over time: Theory or empirics? Scientometrics, 95(3), 1023–1029. Jones, B. F. (2009). The burden of knowledge and the “death of the renaissance man”: Is innovation getting harder?. Review of Economic Studies, 76: 283–317. Jones, B. F. (2010). Age and great invention. Review of Economics and Statistics, 92: 1–14. Jones, B. F., & Weinberg, B. A. (2011). Age dynamics in scientific creativity. Proceedings of the National Academy of Sciences, 108, 18910–18914. Kademani, B. S., Kalyana, V. L., Kumar, V., & Mohan, L. (2005). Nobel laureates: Their publication productivity, collaboration and authorship status. Scientometrics, 62, 261–268. Kahneman, D. (2007). Daniel Kahneman. In G. Linzey & W. M. Runyan (Eds.), A history of psychology in autobiography (Vol. IX., pp. 154– 197). Washington, DC: American Psychological Association. Kalaitzidakis, P., Mamuneas, T. P. & Stengos, T. (2011). An updated ranking of academic journals in economics. Canadian Journal of Economics/Revue canadienne d'économique, 44, 1525–1538. Kalaitzidakis, P., Theofanis, P. M., & Thanasis, S. (2003). Rankings of academic journals and institutions in economics. Journal of the European Economic Association, 1, 1346–1366. Katz, S. J., & Martin, B. R. (1997). What is research collaboration? Research Policy, 26, 1–18. Kearl, J. R., Pope C. L., Whiting B. C. & Wimmer L. T. (1979). A confusion of economists?. American Economic Review 69, 28–37.

247

Kenny, L. W., & Studley, R. E. (1995). Economists’ salaries and lifetime productivity. Southern Economic Journal, 62(2), 382–393. Kestemont, M., Karsdorp, F., & Karsdorp, F. (2014). Mining the twentieth century’s history from the Time Magazine corpus. EACL 2014, 62. Keynes, J. M. (1936). The General Theory of Employment, Interest and Money, Macmillan, London. Kidd, J. S. (1988). The popularization of science: Some basic measurements. Scientometrics, 14(1–2), 127–142. Klahr, D. (2004). Encounters with the force of Herbert A. Simon’. In: Augier, M and March, JG (eds) Models of a Man. Essays in Memory of Herbert A. Simon, pp. 433–49. Cambridge, MA: The MIT Press. Klavans, R., & Boyack, K. W. (2008). Thought leadership: A new indicator for national and institutional comparisons. Scientometrics, 75(2), 239– 250. Kocher, M. G., & Sutter, M. (2001). The institutional concentration of authors in top journals of economics during the last two decades. The Economic Journal, 111(472), 405-421. Kodrzycki, Y. K., & Yu, P. (2006). New approaches to ranking economics journals. The B.E. Journal of Economic Analysis & Policy, 5(1), 1–44. Kolbert, E. (2007). Annals of science: Crash course. Can a seventeen-mile- long collider unlock the universe? The New Yorker. http://www.newyorker.com/reporting/2007/05/14/070514fa_fact_kol bert?currentPage=all. Accessed 31 May 2013. Kousha, K., & Thelwall, M. (2007). The web impact of open access social science research. Library & Information Science Research, 29(4), 495– 507. Kroeber, A. L. (1944). Configurations of culture growth. Berkeley: University of California Press. Kuhn, T. S. (1977). The essential tension. Selected studies in scientific tradition. Chicago: The University of Chicago Press. Kurian, G. T. (2002) The Nobel Scientists: A Biographical Encyclopedia. Amherst, NY: Prometheus Books. Landes, W. M. (2003). The test of time: Does 20th century American art survive? In V. A. Ginsburgh (Ed.), Economics of Art and Culture Invited Papers at the 12th International Conference of the Association of Cultural Economics International (Contributions to Economic

248

Analysis, Volume 260) (pp. 143–164). London: Emerald Group Publishing Limited. Latour, B. & Woolgar, S. (1979). Laboratory life: The social construction of scientific facts. London: Sage. Lawani, S. M. (1986). Some bibliometric correlates of quality in scientific research. Scientometrics, 9(1–2), 13–25. Lehman, H. C. (1953) Age and Achievement. Princeton: Princeton University Press. Leontief, W. (1971). Theoretical assumptions and nonobserved facts. American Economic Journal 61, 1–7. Letierce, J., Passant, A., Decker, S., & Breslin, J. G. (2010). Understanding how Twitter is used to spread scientific messages. In Proceedings of the WebSci10: Extending the Frontiers of Society On-Line, April 26– 27th. Raleigh, NC. Levin, S. G. & Stephan, P. E. (1991). Research Productivity Over the Life Cycle: Evidence for Academic Scientists. American Economic Review, 81: 114–32. Levine-Clark, M., Esther, L., & Gil, E. L. (2008). A comparative citation analysis of Web of Science, Scopus, and Google Scholar. Journal of Business and Finance Librarianship, 14(1), 32–46. Levitt, J. M., & Thelwall, M. (2008). Patterns of annual citation of highly cited articles and the prediction of their citation ranking: A comparison across subjects. Scientometrics, 77(1), 41–60. Li, J., Burnham, J. F., Lemley, T., & Britton, R. M. (2010). Citation analysis: comparison of Web of Science, ScopusTM, SciFinder, and Google Scholar. Journal of Electronic Resources in Medical Libraries, 7(3), 197–217. Lindsey, D. (1980). Production and citation measures in the sociology of science: The problem of multiple authorship. Social Studies of Science 10, 145–62. Lotka, A. J. (1926). The frequency distribution of scientific productivity. Journal of the Washington Academic of Sciences, 16, 317–323. Machin, S. & Oswald A. (1999). Signs of disintegration: A report on UK economics PhDs and ESRC studentship demand. A report commissioned by the Economic and Social Research Council and associated organisations, including the Bank of England and the Government Economic Service, Swindon.

249

Marton, F., Fensham, P., & Chaiklin, S. (1994). A Nobel’s eye view of scientific intuition: Discussions with the Nobel prize-winners in physics, chemistry and medicine (1970–86). International Journal of Science Education, 16, 457–473. Mayer, T. (2004). Comment on “Dry holes in economic research.” Kyklos, 57, 621–625. Mazlish, B. (1982). The quality of “the quality of science”: An evaluation. In M. C. La Follette (ed.), Quality in Science. Cambridge, MA: MIT Press, 48–67. McDowell, J. M. (1982). Obsolescence of knowledge and career publication profiles: Some evidence of differences among fields in costs of interrupted careers. American Economic Review, 72(4), 752–768. Medoff, M. H. (2006). Evidence of a Harvard and Chicago Matthew effect. Journal of Economic Methodology, 13: 485–506. Melguizo, T., & Strober, M. H. (2007). Faculty salaries and the maximization of prestige. Research in Higher Education, 48(6), 633–668. Merton, R. K. (1941). The social function of science. American Journal of Sociology, 46, 622–623. Merton, R. K. (1957). Priorities in scientific discovery: A chapter in the sociology of science. American Sociological Review, 22, 635–659. Merton, R. K. (1968). The Matthew effect in science. Science, 159, 56–63. Merton, R. K. (1969). Behavior patterns of scientists. American Scientist, 57, 1–23. Merton, R. K. (1973). The sociology of science: Theoretical and empirical investigation. Chicago: Chicago University Press. Merton, R. K. (1988). The Matthew effect in science II: Cumulative advantage and the symbolism of intellectual property. Isis, 79, 606– 623. Merton, R. K. (1995). The Thomas theorem and the Matthew effect. Social Forces, 74: 379–424. Meyers, J. (2007). The literary politics of the Nobel Prize. Antioch Review, 65: 214–23. Milojević, S. (2014). Principles of scientific research team formation and evolution. Proceedings of the National Academy of Sciences, 111, 3984–3989.

250

Mingers, J. (2008). Exploring the dynamics of journal citations: Modelling with s-curves. Journal of the Operational Research Society, 59, 1013 –1025. Moore, W. J., Newman, R. J., & Turnbull, G. K. (1998). Do academic salaries decline with seniority? Journal of Labor Economics, 16(2), 352–366. Naroll, R., Benjamin, E. C., Fohl, F. K., Fried, M. J., Hildreth, R. E., & Schaefer, J. M. (1971). Creativity: cross-historical pilot survey. Journal of Cross-Cultural Psychology, 2(2), 181–188. Nature (2013). The maze of impact metrics. Nature, 502, 271. Nederhof, A. J. (2008). Policy impact of bibliometric rankings of research performance of departments and individuals in economics. Scientometrics, 74(1), 163–174. Newman, M. E. J. (2004). Who is the best connected scientist? A study of scientific coauthorship networks. In E. Ben-Naim, H. Frauenfelder, & Z. Toroczkai (Eds.), Complex networks (pp. 337–370). Berlin: Springer. Nicholls, P. T. (1988). Price’s square root law: Empirical validity and relation to Lotka’s law. Information Processing & Management, 24: 469–77. Norrby, E. (2010). Nobel Prizes and life sciences. Hackensack, NJ: World Scientific. Ohba, N., & Nakao, K. (2012). Sleeping beauties in ophthalmology. Scientometrics, 93(2), 253–264. Olry, R. & Haines, D. E. (2012). “Matthew Effect” in Neurosciences. Journal of the History of the Neurosciences, 21: 115–8. Owens, B. (2013). Research assessments: Judgment Day. Nature, 502, 288– 290. Peters, H. P., Brossard, D., de Cheveigné, S., Dunwoody, S., Kaffass, M., Miller, S., et al. (2008). Interaction with the mass media. Science, 321(5886), 204–205. Petersen, A. M., Riccaboni, M., Stanley, H. E. & Pammolli, F. (2012). Persistence and Uncertainty in the Academic Career. Proceedings of the National Academy of Sciences, 109: 5213–8. Piwowar, H. (2013). Value all research output. Nature, 493, 159. Ponomarev, I. V., Williams, D. E., Hackett, C. J., Schnell, J. D., & Haak, L. L. (2014). Predicting highly cited papers: A method for early detection of candidate breakthroughs. Technological Forecasting & Social Change, 81, 49–55.

251

Price, D. D. S. (1963). Little Science, Big Science. New York: Columbia University Press. Price, D. D. S. (1976). A general theory of bibliometric and other cumulative advantage processes. Journal of the Association for Information Science and Technology, 27, 292–306. Priem, J. (2013). Beyond the paper. Nature, 495, 437–440. Priem, J., & Hemminger, B. M. (2010). Scientometrics 2.0: Toward new metrics of scholarly impact on the social web. First Monday, 15, 7. Prigogine, I., & Stengers, I. (1984). Order out of chaos: Man’s new dialogue with nature. Toronto: Bantam Books. Putnam, L. L. (2009). Symbolic capital and academic fields: An alternative discourse on journal rankings. Management Communication Quarterly, 23, 127–134. Rablen, M. D., & Oswald, A. J. (2008). Mortality and immortality: The Nobel Prize as an experiment into the effect of status upon longevity. Journal of Health Economics, 27(6), 1462–1471. Radicchi, F., Fortunato, S., & Castellano, C. (2008). Universality of citation distributions: Toward and objective measure of scientific impact. Proceedings of the National Academy of Sciences, 105(45), 17268– 17272. Redner, S. (1998). How popular is your paper? An empirical study of the citation distribution. The European Physical Journal B - Condensed Matter and Complex Systems, 4(2), 131–134. Redner, S. (2005). Citation statistics from 110 years of physical review. Physics Today, 58(6), 49–53. RePEc (2013a). Doctoral advisors and programs. April 29, http://blog.repec.org/2013/04/29/proposed-changes-to-repec- rankings-up-for-vote (last accessed 30 July 2013). RePEc (2013b). Regarding Citation Counts. April 29, http://blog.repec.org/2013/04/29/ proposed-changes-to-repec- rankings-up-for-vote (last accessed 30 July 2013). Reskin, B. F. (1977). Scientific productivity and the reward structure of science. American Sociological Review, 42(3), 491–504. Rosen, S. (1981). The economics of superstars. American Economic Review 71, 845–58. Rossiter, M. W. (1993). The Matthew Matilda effect in science. Social Studies of Science, 23: 325–41.

252

Roth, C., Wu, J., Lozano, S. (2012). Assessing impact and quality from local dynamics of citation networks. Journal of Informetrics, 6(1), 111–120. Rothenberg, A. (2005). Family background and genius II: Nobel laureates in science. Canadian Journal of Psychiatry, 50, 918–925. Rowland, F. S. (1993/1995). President’s lecture: The need for scientific communication with the public. Science, 260(5114), 1571–1576. Rubenstein, L. D. (2012). Using TED Talks to inspire thoughtful practice. The Teacher Educator, 47(4), 261–267. Samuels, W. J. (1980). Economics as a science and its relation to policy: The example of free trade. Journal of Economic Issues 14, 163–85. Samuelson, P. A. (2004). Paul A. Samuelson. In W. Breit & B. T. Hirsch (Eds.), Lives of the laureates: Eighteen Nobel Economists (pp. 49–64). Cambridge, MA: MIT Press. Sangwal, K. (2013). Growth dynamics of citations of cumulative papers of individual authors according to progressive nucleation mechanism: Concept of citation acceleration. Information Processing & Management, 49(4), 757–772. Sapienza, P. & Zingales L. (2013). Economic experts vs. average Americans. American Economic Review 103, 636–42. Schettino, F., Sterlacchini, A. & Venturini, F. (2013). Inventive productivity and patent quality: Evidence from Italian inventors. Journal of Policy Modeling, 35: 1043–56. Schumpeter, J. A. (1934/1983). The theory of economic development: An inquiry into profits, capital, credit, interest, and the business cycle. New Brunswick, NJ: Transaction Publishers. Schweber, S. S. (2003). Quantum field theory: From QED to the standard model. In M. J. Nye (Ed.), The Cambridge History of Science. Volume 5. The modern physical and mathematical sciences (pp. 375–393). Cambridge: Cambridge University Press. Schweber, S. S. (2008). Einstein and Oppenheimer: The meaning of genius. Cambridge, MA: Harvard University Press. Scott, L. C. & Mitias P. M. (1996). Trends in ranking of economics departments in the U.S.: An update. Economic Inquiry 34, 378–400. Seglen, P. O. (1992). The skewness of science. Journal of the American Society for Information Science, 43: 628–38.

253

Shaman, J., Solomon, S., Colwell, R. R., & Field, C. B. (2013). Fostering advances in interdisciplinary climate science. Proceedings of the National Academy of Sciences, 110, 3653–3655. Shapiro, F. R. (1992). Origins of bibliometrics, citation indexing and citation analysis: The neglected legal literature. Journal of the American Society for Information Science, 43(5), 337–339. Shavinina, L. V. (2004). Explaining high abilities of Nobel laureates. High Ability Studies, 15, 243–254. Simmons, Jr., L. M., & West, G. B. (1981). Physics, Philosophy, leadership, policy: An interview with Peter Carruthers. Los Alamos Science, 2, 133–143. Simon, H. A. (1955). On a class of skew distribution functions. Biometrika 42, 425–40. Simon, H. A. (1996) Models of My Life. Cambridge, MA: The MIT Press. Simon, H. A. (2008). Colloquium with H. A. Simon. In H. A. Simon, M. Egidi, R. Marris, & R. Viale (Eds.), Economics, bounded rationality and the cognitive revolution (pp. 8–36). Edward Elgar: Cheltenham. Simonton, D. K. (1975a). Sociocultural context of individual creativity: A transhistorical time-series analysis. Journal of Personality and Social Psychology, 32(6), 1119–1133. Simonton, D. K. (1975b). Age and literary creativity: A cross-cultural and transhistorical survey. Journal of Cross-Cultural Psychology, 6(3), 259–277. Simonton, D. K. (1976a). Philosophical eminence, beliefs, and zeitgeist: An individual-generational analysis. Journal of Personality and Social Psychology, 34(4), 630–640. Simonton, D. K. (1976b). The causal relation between war and scientific discovery: An exploratory crossnational analysis. Journal of Cross- Cultural Psychology, 7(2), 133–144. Simonton, D. K. (1976c). Biographical determinants of achieved eminence: A multivariate approach to the Cox data. Journal of Personality and Social Psychology, 33(2), 218–226. Simonton, D. K. (1978). The eminent genius in history: The critical role of creative development. Gifted Child Quarterly, 22(2), 187–195. Simonton, D. K. (1980). Thematic fame, melodic originality, and musical zeitgeist: A biographical and transhistorical content analysis. Journal of Personality and Social Psychology, 38(6), 972–983.

254

Simonton, D. K. (1984). Artistic creativity and interpersonal relationships across and within generations. Journal of Personality and Social Psychology, 46(6), 1273–1286. Simonton, D. K. (1988a). Age and outstanding achievement: What do we know. Psychological Bulletin, 104: 251–67. Simonton, D. K. (1988b) Scientific Genius: A Psychology of Science. Cambridge: Cambridge University Press. Simonton, D. K. (1992). Age and achievement. In: Albert, RS (ed.) Genius & Eminence, pp. 77–89. Oxford: Pergamon Press. Simonton, D. K. (2013). Scientific genius is extinct. Nature, 493(7434), 602. Simonton, D. K. (2014a). Creative performance, expertise acquisition, individual differences, and developmental antecedents: An integrative research approach. Intelligence, 45, 66–73. Simonton, D. K. (Ed.). (2014b). The Wiley handbook of genius. Oxford: Wiley Blackwell. Sorokin, P. A. (1957). Social and cultural dynamics. Boston: Porter Sargent Publisher. Stack S. (2004). Gender, children and research productivity? Research in Higher Education, 45, 891–920. Stadelmann, D. & Torgler, B. (2013). Bounded Rationality and Voting Decisions over 160 Years: Voter behavior and increasing complexity in decision-making. PLoS One, 8: e84078. Stallings, J., Vance, E., Yang, J., Vannier, M. W., Liang, J., Pang, L., et al. (2013). Determining scientific impact using a collaboration index. Proceedings of the National Academy of Sciences of the United States of America, 110, 9680–9685. Stanovich, K. E. (1986). Matthew Effects in reading: Some consequences of individual differences in the acquisition of literacy. Reading Research Quarterly, 21: 360–407. Stephan, P. E. (2012) How Economics Shapes Science. Cambridge, MA: Harvard University Press. Stephan, P. E., & Levin, S. G. (1993). Age and the Nobel Prize revisited. Scientometrics, 28, 387–399. Stigler, G. J., Stigler, S. M., & Friedland, C. (1995). The journals of economics. Journal of Political Economy, 103, 331–359.

255

Stokols, D., Hall, K. L., Taylor, B. K., & Moser, R. P. (2008). The science of team science: Overview of the field and introduction to the supplement. American Journal of Preventive Medicine, 25, S77–S89. Strevens, M. (2006). The role of the Matthew effect in science. Studies in History and Philosophy of Science, 37: 159–70. Sugimoto, C. R., & Thelwall, M. (2013). Scholars on soap boxes: Science communication and dissemination in TED videos. Journal of the American Society for Information Science and Technology, 64(4), 663– 674. Summers, L. H. (2000). International financial crises: Causes, preventions and cures. American Economic Review 90, 1–16. The Economist (1997). The puzzling failure of economics. The Economist, August 23, p. 13. The Economist (2000). Economics forum: The future of economics. The Economist, March 4, p. 90. The Economist (2015). Influential economists: That ranking. http://www.economist.com/blogs/freeexchange/2015/01/influential- economists (last accessed 12 February 2015). Thelwall, M. (2008). Bibliometrics to webometrics. Journal of Information Science, 34(4), 605–621. Thelwall, M., & Harries, G. (2004). Do the web sites of higher rated scholars have significantly more online impact? Journal of the American Society for Information Science and Technology, 55(2), 149–159. Thelwall, M., & Price, L. (2003). Disciplinary differences in academic web presence: A statistical study of the UK. Libri, 53, 242–253. Tol, R. S. J. (2009). The Matthew effect defined and tested for the 100 most prolific economists. Journal of the American Society for Information Science and Technology, 60: 420–6. Torgler, B. & Piatti, M. (2013). A century of American Economic Review: Insights on critical factors in journal publishing. Basingstoke, UK: Palgrave/MacMillan. Trajtenberg, M. (1990). A penny for your quotes: Patent citations and the value of innovations. RAND Journal of Economics 21, 172–87. Turchin, P. (2003). Historical dynamics: Why states rise and fall. Princeton: Princeton University Press. Turchin, P., & Nefedov, S. A. (2009). Secular cycles. Princeton: Princeton University Press.

256

Uzzi, B., Mukherjee, S., Stringer, M., & Jones, B. (2013). Atypical combinations and scientific impact. Science, 342, 468–472. Van Bergeijk, P. A. G., Bovenberg, A. L., van Damme, E. E. C., & van Sinderen, J. (1997). Economic science and practice: The roles of academic economists and policy-makers. Cheltenham: Edward Elgar. van Dalen, H. J. & Klamer A. (1997). Blood is thicker than water: Economists and the Tinbergen legacy. In P. A. G. van Bergeijk, A. L. Bovenberg, E. E. C. van Damme and J. van Sinderen, eds, Economic Science and Practice: The Roles of Academic Economists and Policymakers, Edward Elgar, Cheltenham, pp. 60–91. van Dalen, H. P. & Klamer A. (2005). Is science a case of wasteful competition?. Kyklos 58, 395–414. van Dalen, H. P. (1999). The golden age of Nobel economics. The American Economist, 43, 19–35. Van de Ven, A. H., & Johnson, P. E. (2006). Knowledge for theory and practice. Academy of Management Review, 31(4), 802–821. Van Looy, B., Ranga, M., Callaert, J., Debakere, K. & Zimmermann, E. (2004). Combining entrepreneurial and scientific performance in academia: Towards a compounded and reciprocal Matthew effect?. Research Policy, 33: 425–41. Van Noorden, R. (2010). A profusion of measures. Nature, 466, 864–866. Van Raan, A. F. J. (1997). Scientometrics: State-of-the-art. Scientometrics, 38(1), 205–218. van Rijnsoever, F. J., & Hessels, L. (2011). Factors associated with disciplinary and interdisciplinary research collaboration. Research Policy, 40, 463–472. Walberg, H. J. & Tsai, S. L. (1983). Matthew effects in education. American Educational Research Journal, 20: 359–73. Waldinger, F. (2012). Peer effects in science: Evidence from the dismissal of scientists in Nazi Germany. Review of Economic Studies, 79(2), 838– 861. Wall, H. (2009). Journal rankings in economics: handle with care. Working paper 2009-014A, Federal Reserve Bank of St. Louis. Wallace, M. L., Larivière, V., & Gingras, Y. (2009). Modeling a century of citation distributions. Journal of Informetrics, 3(4), 296–303. Wang, D., Song, C., & Barabási, A.-L. (2013). Quantifying long-term scientific impact. Science, 342(6154), 127–132.

257

Wang, J. (2014). Unpacking the Matthew effect in citations. Journal of Informetrics, 8: 329–39. Watson, J. D. (2007). Avoid Boring People: And other Lessons from a Life of Science. New York: Alfred A. Knopf. Watts, C., & Gilbert, N. (2011). Does cumulative advantage affect collective learning in science? An agent-based simulation. Scientometrics. 89(1), 437–463. Whitfield, J. (2008). Group theory. Nature, 455, 720–723. Willems, J. (2003). Bringing down the barriers: Public communication should be part of common scientific practice. Nature, 422, 470. Wilson, E. O. (1998). Consilience: The unity of knowledge. New York: Vintage Books. Wilson, E. O. (2013), Letters to a Young Scientist, Liveright Publishing Corporation, New York. Wray, K. B. (2006). Scientific authorship in the age of collaborative research. Studies in History and Philosophy of Science, 37, 505–514. Wu, Q. (2010). The w-index: A measure to assess scientific impact by focusing on widely cited papers. Journal of the American Society for Information Science and Technology 61, 609–14 Wuchty, S., Jones, B. F., & Uzzi, B. (2007). The increasing dominance of teams in production of knowledge. Science, 316(5827), 1036–1039. Yukawa, H. (1978). Creativity and intuition: A physicist looks at east and west. Tokyo, New York: Kodansha International. Zuckerman, H. (1992). The proliferation of prizes: Nobel complements and Nobel surrogates in the reward system of science. Theoretical Medicine, 13, 217–231. Zuckerman, H. (1996) Scientific Elites: Nobel Laureates in the United States. New Brunswick: Transaction Publishers.

258

Appendices: Statements of Authors Contributions

259

Chapter 2: The first cut is the deepest: Repeated interactions of co-authorship and academic productivity in Nobel laureate teams Statement of Contribution of Co-Authors for

Thesis by Published Paper

The authors listed below have certified* that: 1. they meet the criteria for authorship in that they have participated in the conception, execution, or interpretation, of at least that part of the publication in their field of expertise; 2. they take public responsibility for their part of the publication, except for the responsible author who accepts overall responsibility for the publication; 3. there are no other authors of the publication according to these criteria; 4. potential conflicts of interest have been disclosed to (a) granting bodies, (b) the editor or publisher of journals or other publications, and (c) the head of the responsible academic unit, and 5. they agree to the use of the publication in the student’s thesis and its publication on the QUT ePrints database consistent with any limitations set by publisher requirements. In the case of this chapter:

The first cut is the deepest: Repeated interactions of co-authorship and academic productivity in Nobel laureate teams

Scientometrics (2016), 106(2), 509-524.

Contributor Statement of contribution Ho Fai Chan Has equally contributed to all aspects of this paper, including

research, analysis and manuscript writing. 6 March 2017 Has equally contributed to all aspects of this paper, including Ali Sina Önder research, analysis and manuscript writing. Has equally contributed to all aspects of this paper, including Benno Torgler research, analysis and manuscript writing.

Principal Supervisor Confirmation

I have sighted email or other correspondence from all Co-authors confirming their certifying authorship.

QUT Verified Signature

Benno Torgler 6/03/2017 ______

Name Signature Date

260

Chapter 3: Do Nobel laureates change their patterns of collaboration following prize reception?

Statement of Contribution of Co-Authors for

Thesis by Published Paper

Do Nobel laureates change their patterns of collaboration following prize reception?

Scientometrics (2015), 105(3), 2215-2235.

Contributor Statement of contribution Ho Fai Chan Has equally contributed to all aspects of this paper, including

Principal Supervisor Confirmation

I have sighted email or other correspondence from all Co-authors confirming their certifying authorship.

QUT Verified Signature

Benno Torgler 6/03/2017 ______

Name Signature Date

261

Chapter 4: Awards before and after the Nobel Prize: a Matthew effect and/or a ticket to one's own funeral?

Statement of Contribution of Co-Authors for

Thesis by Published Paper

Awards before and after the Nobel Prize: a Matthew effect and/or a ticket to one's own funeral?

Research Evaluation (2014), 23(3), 210-220.

Contributor Statement of contribution Ho Fai Chan Contributed in the data collection, analysis and

manuscript writing. 6 March 2017 Contributed in the data collection, analysis and Laura Gleeson manuscript writing. Directed the research, contributed to data analysis and Benno Torgler manuscript writing.

Principal Supervisor Confirmation

I have sighted email or other correspondence from all Co-authors confirming their certifying authorship. QUT Verified Signature

Benno Torgler 6/03/2017 ______

Name Signature Date

262

Chapter 5: The implications of educational and methodological background for the career success of Nobel Laureates: an investigation of major awards

Statement of Contribution of Co-Authors for

Thesis by Published Paper

The implications of educational and methodological background for the career success of Nobel Laureates: an investigation of major awards

Scientometrics (2015), 102(1), 847-863.

Contributor Statement of contribution Ho Fai Chan Has equally contributed to all aspects of this paper,

including research, analysis and writing. 6 March 2017 Has equally contributed to all aspects of this paper, Benno Torgler including research, analysis and writing.

Principal Supervisor Confirmation

I have sighted email or other correspondence from all Co-authors confirming their certifying authorship.

QUT Verified Signature

Benno Torgler 6/03/2017 ______

Name Signature Date

263

Chapter 6: External influence as an indicator of scholarly importance

Statement of Contribution of Co-Authors for

Thesis by Published Paper

External influence as an indicator of scholarly importance

CESifo Economic Studies (2016), 62(1), 170-195.

Contributor Statement of contribution Ho Fai Chan Contributed to the data capture, data analysis and

manuscript writing. 6 March 2017 Bruno Frey Contributed to the manuscript writing. Jana Gallus Contributed to the manuscript writing. Established methodology for data collection, Markus Schaffner contributed to manuscript writing. Directed the research and contributed to all aspects of Torgler Benno the paper. Stephen Whyte Contributed to the data capture.

Principal Supervisor Confirmation

I have sighted email or other correspondence from all Co-authors confirming their certifying authorship.

QUT Verified Signature

Benno Torgler 6/03/2017 ______

Name Signature Date

264

Chapter 7: Do the best scholars attract the highest speaking fees? An exploration of internal and external influence

Statement of Contribution of Co-Authors for

Thesis by Published Paper

Do the best scholars attract the highest speaking fees? An exploration of internal and external influence

Scientometrics (2014), 101(1), pp. 793-817.

Contributor Statement of contribution Ho Fai Chan Contributed to the data capture, data analysis and manuscript

writing. 6 March 2017 Bruno Frey Contributed to the manuscript writing. Jana Gallus Contributed to the manuscript writing. Established methodology for data collection and capture, Markus Schaffner contributed to manuscript writing. Directed the research and contributed to all aspects of the Torgler Benno paper Stephen Whyte Contributed to the data capture.

Principal Supervisor Confirmation

I have sighted email or other correspondence from all Co-authors confirming their certifying authorship. QUT Verified Signature

Benno Torgler 6/03/2017 ______

Name Signature Date

265

Chapter 8: The inner quality of an article: Will time tell?

Statement of Contribution of Co-Authors for

Thesis by Published Paper

The inner quality of an article: Will time tell?

Scientometrics (2015), 104(1), 19-41.

Contributor Statement of contribution Ho Fai Chan Conducted the data collection, data analysis and interpretation, and

manuscript writing. 6 March 2017 Malka Guillot Contributed to the data analysis and manuscript writing. Lionel Page Contributed the original idea and reviewed the manuscript. Directed the research, contributed to data analysis and manuscript Benno Torgler writing.

Principal Supervisor Confirmation

I have sighted email or other correspondence from all Co-authors confirming their certifying authorship.

QUT Verified Signature

Benno Torgler 6/03/2017 ______

Name Signature Date

266

Chapter 9: Do great minds appear in batches?

Statement of Contribution of Co-Authors for

Thesis by Published Paper

Do great minds appear in batches?

Scientometrics (2015), 104(2), pp. 475-488.

Contributor Statement of contribution Ho Fai Chan Has equally contributed to all aspects of this paper,

including research, analysis and writing. 6 March 2017 Has equally contributed to all aspects of this paper, Benno Torgler including research, analysis and writing.

Principal Supervisor Confirmation

I have sighted email or other correspondence from all Co-authors confirming their certifying authorship.

QUT Verified Signature

Benno Torgler 6/03/2017 ______

Name Signature Date

267