Human capital spillover effects from star scientists to co-author
generation of innovative ideas
and the moderating role of co-author age
MSc in Business Administration: Strategy track
MASTER THESIS
Author: Renars Kukuks 11801336
Supervisor:
Dr. N.E. Betancourt
Date: 22/06/2018
Statement of Originality
This document is written by Renars Kukuks who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Table of Contents
1. Introduction ... 5
2. Literature Review ... 7
2.1 The Role of Stars and Co-authors ... 7
2.2 Stars and Human Capital Spillovers to Co-authors ... 8
2.3 Stars and Knowledge Sharing Motivation ... 10
2.4 Co-authors and Knowledge Intake Motivation ... 13
2.5 Co-authors and Variance Between Spillover Effects ... 15
2.6 Co-authors and Influence by Social Comparisons ... 16
2.7 Co-author’s Age as a Proxy for Social Comparisons ... 18
3. Methodology ... 21
3.1 Population ... 21
3.2 Sample ... 22
3.3 Data Collection Process ... 22
3.4 Measurements ... 26
3.3.1 Dependent Variable - Generation of Innovative Ideas ... 26
3.3.2 Independent Variable – Co-author post star ... 27
3.3.3 Moderating Variable – Academic age ... 27
3.3.4 Control Variables ... 28 4. Results ... 29 4.1 Descriptive Statistics ... 29 4.2 Regression Analysis ... 30 4.2.1 Hypothesis 1 ... 31 4.2.2 Hypothesis 2 ... 32 4.3 Robustness ... 33 5. Discussion ... 38
6. Limitations and Suggestions for Further Research ... 41
7. Conclusion ... 43
Abstract
Given the rise in importance of innovation in knowledge driven economy, there is a growing body of research focused on star scientists and the peer effects they produce on their collaborators. There is an agreement that star scientists make their co-authors more productive in generation of innovative ideas, but not all co-authors benefit equally by working with a star. Existing literature indicates that characteristics of a star determine when do peer effects happen. This study proposes that characteristics of co-authors also determine when do stars make them more productive. It proposes that co-authors experience different gaps resulting from social comparisons post working with a star. Bigger gaps motivate co-authors to be more productive. By using historical data of the star population who participated in the Manhattan Project (1939 – 1945) during World War II, this research systematically sampled 30 stars and researched the patenting output of their respective co-authors before and after the collaboration with a star. The social comparison gaps were measured by observing the moderating effect of co-author career age. The findings show that co-authors who worked with stars became more productive patent producers in comparison to co-authors who never worked with stars. Moreover, the moderating influence of co-authors age proved significant. Younger co-authors who worked with a star produced more patents in comparison to older co-authors. This study added evidence to the literature by showing how innovative productivity increases right after working with a star. It also adds to the literature by showing that peer effects happen not only due to star characteristics, but also due to co-author characteristics.
1. Introduction
Innovation is at the heart of economic growth. It is driven by the production of scientific and technological ideas which stem from the knowledge residing within individuals. Thus, individuals are at the core of innovation advancement (Almeida & Kogut, 1999; Howells, 2002; Lynne G. Zucker & Darby, 1996). However, not all individuals are equally capable of generating innovative ideas. In fact, a small proportion of all scientists account for most of the output. Organizations recognise this and compete on a global scale to attract the most talented individuals (Kapur & McHale, 2005). Research has identified such people as stars. The role of these people is substantial and much has been researched on their influence on organizational performance and individual scientific output (Azoulay, Zivin, S, & Wang, 2010; Hess & Rothaermel, 2012; Oettl, 2012; Waldinger, 2012a; L.G. Zucker, Darby, & Armstrong, 2002). However, technological innovation is not solely achieved by stars. It is an achievement of collaborative efforts (Hess & Rothaermel, 2012). Moreover, when stars work together with their collaborators, they produce human capital spillovers, making them more productive “innovators” afterwards (Azoulay, Fons-Rosen, & Zivin, 2015; Moser, Voena, & Waldinger, 2014; Waldinger, 2012a, 2016). This illuminates the importance on the role of close collaborators, as many of them contribute to organizational innovation. But despite the importance of spillover effects to collaborators, only a few studies have taken them as the unit of analysis in academia. There is a general agreement that stars make co-authors more productive post collaboration (Azoulay et al., 2010; Borjas & Doran, 2012). Still, what remains unknown is when is this actually happening. The spillover effects are unevenly distributed across co-authors (Oettl, 2012). It points to the fact that not all co-authors are equally benefitting by working with stars. This study aims to address this issue by examining co-authors of star scientists who participated in the Manhattan Project from 1939 to 1945, which produced atomic bombs.
Empirical evidence shows us that human capital spillover effects are also affected by the characteristics of a star. Oettl (2012) finds that the helpful and supportive stars make their peers more productive than non-supportive stars. Grigoriou & Rothaermel (2014) claim that networking-savvy and social stars also prove to make others more productive than non-social stars. Building from these observations, I propose that star motivation also plays a role. The novel knowledge which stars attained in Manhattan Project should motivate them to share it, and therefore give co-authors the input to be more productive.
For explanations when would the human capital spillovers happen, I refer to the theory of social comparisons. Literature indicates that individuals in scientific departments compare their achievements to their peers to self-validate themselves (Tartari, Perkmann, & Salter, 2014). As a result, there is a perceived gap between them and their peers. As a response, individuals “copy” their peer behaviour to lessen this gap. This effect is even stronger when the peers are perceived of prestigious social status (Stuart & Ding, 2006). This phenomenon implies us to assume that star and co-author collaborations would play out similarly. But what is impossible is that all co-authors will experience the same perceived gap between them and the star. Some would have to produce more scientific work to lessen the gap than others. I hypothesise that this is a potential source of the unevenly distributed human capital spillover effects. Although, it is challenging to measure such circumstance. I propose that age of the co-authors could act as a proxy for the gap size resulting from social comparisons. Younger scientists have not had the same amount of time in their careers as older ones. Therefore, when a young co-author compares himself with a star, he may experience larger gap than an older co-author. As a result, the younger co-author could be more motivated to produce innovative
co-authors post working with a star and generation of innovative ideas. Whether significant or not, the findings should provide a partial answer to the productivity variance problem. Prior studies researched peer effects by measuring productivity decrease when star passes away or leaves geographical proximity. This study goes against the traditional method and shows productivity change of co-authors in the period after working with a star. The results indicate what happens when the star is still within co-authors social network. Lastly, prior studies concluded that peer effects happen when stars are of relational and collaborative character. This study adds the role of co-author age to the list of reasons when spillover effects happen.
2. Literature Review
2.1 The Role of Stars and Co-authors
Star performers are individuals who produce significantly more output than their respective peers. These individuals exist across multiple industries such as sports, entertainment, and industrial engineering. In science, stars are known as star scientists who are highly accomplished experts holding PhD degrees (Hess & Rothaermel, 2012). They account for approximately 5% of the total labour market within a particular science area. These have been classified as either highly cited or highly funded scientists, or simply being winners of prizes in their respective fields (Azoulay et al., 2010). They also tend to lead the direction of knowledge in their scientific domains (Azoulay et al., 2015).
However, what is interesting is that besides directing the development of knowledge and producing knowledge, stars are unlikely to take the role of innovative idea production. Innovative ideas have been associated with either patent publishing or founding new enterprises which products or services build on the innovative idea (Haeussler & Colyvas, 2011). Studies have examined what roles in science do disproportionately productive individuals take and in what activities the rest of scientists indulge in (Hess & Rothaermel, 2012; Phillips & Zuckerman, 2001). When given both scientific and commercial opportunity,
scientists take either a scientific role (producing knowledge or tacit research activities) or an applied science role (producing codified research activities) (Furukawa & Goto, 2006). They propose this happens due to U-shaped relationship between status and conformity. Stars are more likely to focus on scientific knowledge production, pursuing research streams that are riskier but offer higher individual rewards (low conformity). This is due to their talented capabilities and due to their social status in the organisation. High status individuals feel secure in their positions and are confident that their behaviour will not cost them their job (Phillips & Zuckerman, 2001). However, non-stars are middle status individuals, holding a lower position in organisations, having high conformity, and thus are not experiencing the same privilege. Given their talent and aspirations for career progression, they tend to be more careful and engage in codified research activities that are in line with organisation goals. These tend to be applied science activities that build innovative ideas on already existing knowledge. Therefore, when given scientific and commercial opportunities at the same time, non-stars tend to take applied science roles, whereas stars tend to take scientific roles.
2.2 Stars and Human Capital Spillovers to Co-authors
The roles tend to be different but also complementary. It has been observed that in pharmaceutical industry stars scientists act as knowledge gatherers from outside-in and input this knowledge across the organisation (Furukawa & Goto, 2006). The people that these stars work in teams with, namely co-authors, take the knowledge after the collaboration and codify it, exploiting it into patents. Firm’s intellectual human capital, specifically patenting, is directly influenced by non-stars, and stars simply act as guidance for knowledge (Rothaermel & Hess, 2007). Thus, stars have a positive effect on the number of patent applications of their
co-Human capital is defined as “the knowledge, skills, and abilities residing with and utilized by individuals” (Subramaniam & Youndt, 2005). The spillover of human capital is an external effect of an individual’s human capital. The effect happens when one’s human capital affects the productivity of another (Oettl, 2012). Oettl (2012) finds that co-author knowledge production activity changes after death of a star they have once worked with. While quantity of papers published remained the same, citation amount decreased. Other studies find that paper publishing rate of co-authors decreases by 5 to 8% after star’s death. Likely cause of it is that co-authors lose a source of intellectual ideas and cannot acquire new knowledge on their own (Azoulay et al., 2010). Regarding the production of innovative ideas, there is empirical evidence that co-authors output increases after collaborating with stars. Furukawa & Goto (2006) found that co-authors who have worked with stars in Japanese Pharmaceutical companies, increase the number of patents published after the collaboration. But why would the human capital spillovers happen in the first place?
To why these spillover effects happen, the conventional agreement is that this happens due to stars acting as a source of knowledge input to co-authors and that close collaborations are the most effective channels for knowledge transfer (Grigoriou & Rothaermel, 2014).
Firstly, stars collect and absorb academic knowledge and serve as channels of knowledge flows from external sources (Furukawa & Goto, 2006; Grigoriou & Rothaermel, 2014). Through collaboration, they add new knowledge to co-author’s human capital which they would not be able to acquire and assimilate on their own. At the essence of innovation is recombination of existing knowledge (Galunic & Rodan, 1998; Murray & O’Mahony, 2007; Zahra & George, 2002). Individuals acquire new knowledge and combine it with their existing one. As a result, individuals have a larger perspective and thus know more options of what is possible. Stars add more knowledge to the co-author’s human capital and co-authors have more
knowledge to recombine as a result. Therefore, human capital spillovers from stars make co-authors more productive innovative idea producers.
Secondly, closely tied and frequent team collaboration is an effective facilitator of transferring knowledge from one agent to another (Groysberg, Lee, & Nanda, 2008). When co-authors work with stars, they are not only given access to unique information, but also the right circumstances for effective human capital spillover. Such conditions are known to be correlated with increased patent output (Schiffauerova & Beaudry, 2011).
2.3 Stars and Knowledge Sharing Motivation
It is clear that post working with a star, co-authors are more productive. However, the characteristics and conditions of the star can affect the spillover effect as well. Oettl (2012) researched how the “helpfulness” of a star influences this. Not all stars produce the same spillover effect. Oettl (2012) concludes that it depends how helpful stars are to their co-working peers. Furthermore, research shows that the position a star occupies in an organization’s network also matters. Stars with strong collaboration skills, named “relational stars”, have central network positions with multiple knowledge flow inputs. This makes them able to accumulate and recombine more knowledge. Such stars spillover knowledge to their peers more effectively than non-relational stars (Grigoriou & Rothaermel, 2014). Lastly, expertise of the star also matters. Co-authors who have similar expertise to that of a star, experience bigger productivity loss after the death of a star than co-authors who have more differentiated expertise (Azoulay et al., 2015; Jaravel, Petkova, & Bell, 2016a).
Despite the research on how star characteristics influence the spillover effect, the role of their motivation to do so is overlooked. For an individual to share knowledge, he must be
In science, whether the knowledge is novel or non-novel, should make the difference in the star’s perception of its value. Novel knowledge has high scientific or commercial benefits and very low amount of people possess it. Stars who possess novel knowledge should be motivated to share it. Therefore, human capital spillover effects to co-authors should happen because the stars are more likely to extensively share novel knowledge. However, as Ipe (2003) states, there need to be the right incentives for the knowledge sharing to happen. To discuss the incentives that stars may have, I look into the knowledge sharing motivation literature.
Motivation to share knowledge has been researched in knowledge work industries. Hendriks (1999) looked into what motivates individuals to share knowledge across information technology systems. To build theoretical background for the motivation factors, Hendriks (1999) used the Herzberg’s (1987) two-factor motivational theory, arguing that it is the best fit due to its wide acceptance, empirical evidence, and its fit to knowledge work. The two-factor theory states that behaviour is maintained by hygiene factors and motivated by motivating factors. Hygiene factors (salary, status, company policy, interpersonal relations) do not motivate when present, but decrease motivation when absent. In contrast, motivating factors (achievement, responsibility, recognition, promotional opportunities, challenge of work) when present, increase motivated behaviour, and decrease it significantly when absent (Herzberg, 1968). Hendriks (1999) also distinguishes between motivation of knowledge owners who transmit knowledge and knowledge reconstructors who absorb knowledge. In the context of this study, owners are stars and recombinators are co-authors. Knowledge owners share knowledge due to their aspirations for recognition for their work, due to promotional opportunities, and due to sense of responsibility. They also share knowledge in expectation for reciprocity, as they expect the other party to share their knowledge in return. On the other side, knowledge reconstructors are motivated to take in the shared knowledge due to the challenge of work, promotional opportunities and sense of achievement.
As each agent has their own motivating factors, these factors should be satisfied for knowledge sharing to happen, and therefore human capital spillovers to happen. These motivators must be present for novel knowledge to be shared. Stars should be motivated to share their novel knowledge with their co-authors because it can yield them recognition. Firstly, stars might share it to be admired for the novelty that they know. Secondly, as stars strive for recognition, it is important that all scientists invest their best efforts in the co-work with stars. For co-authors to do so, they must know all that is relevant. Sharing novel knowledge can also provide promotional opportunities. After working with a star, co-authors might produce knowledge or innovative ideas referring back to their earlier work with stars. This gives star promotion and therefore stars might be motivated to give their co-authors the knowledge to be able to perform later. Finally, when star scientists hold novel knowledge, they are one of the few people who possess it, if not the only ones. As scientists who most likely care about the advancement of their respective fields, stars might feel the sense of responsibility to pass along their knowledge, therefore caring about their co-authors becoming more productive and advancing their knowledge into practical applications.
I propose that co-authors will have higher output of innovative ideas not only due to unique knowledge inputs, knowledge recombination, and close collaboration but also due to star motivation to share novel knowledge with co-authors.
2.4 Co-authors and Knowledge Intake Motivation
There is evidence that if an individual has an opportunity to work together with a star, the individual will get “psyched-up” about it (Boggiano, Pittman, & Shantz, 1992; Flynn & Amanatullah, 2012). Studies on this have looked in two directions. When both actors depend on each other for the task outcome and when the task outcome is independent. For the mutually dependent outcome, Kohler (1926) found that when two individuals are assigned a task together, the high-performance actor “inspires” the other to perform better. This effect has been studied further and has yielded empirical evidence to the so-called discrepancy effect. When two people perform a task together and they have great discrepancies in their performance, the less performing individual experiences motivation gains (Hertel, Kerr, & Messé, 2000; Messé, Hertel, Kerr, Lount, & Park, 2002).
Regarding independent task outcomes, Flynn & Amanatullah (2012) studied similar phenomena but in a different context. They introduced the role of social status (when status depends on past performance) and task outcome independence between both of the actors. They looked at how does the presence of a high-status individual affect a lower status individual when working independently on very similar tasks. High-status performers have positive influence on the independent task performance of the other individual. The mechanism behind this is that high-status actors have inflated performance expectations. They raise the bar of what they expect for the people working next to them. As a response, the co-actors elevate their motivation to match the expectations of the high-status performer. Flynn & Amanatullah (2012) infer that individuals set higher goals for themselves because they see high-status performers as reference points to their aspirations and their high expectations as “anchors”. Research on scientific collaborations also shows that scientists are more motivated to collaborate when the other partner has a “prestige” to their name. If a scientist can show he has
worked with eminent colleagues, it increases his chances for project funding and journals accepting his papers in the future (Noriko, Paul, Seung-Lye, & H, 2003).
The two research areas of dependent task outcomes and independent outcomes, have drawn their conclusions either from laboratory experiments or observations of single task performances such as sports activities and certain problem solving. In contrast, scientific work is a combination of multiple tasks of high complexity. However, as the knowledge sharing motivations are more psychological than task focused, I assume they will be present in a co-author working with a star as well. Co-co-authors should be motivated to intake the knowledge from stars if there are challenges of work, promotional opportunities, and sense of achievement. As co-authors experience discrepancies between their and star performance, they may experience motivation gains to even the differences out. Because of possible motivation gains from the high reference anchor that stars establish, I assume that co-authors will be motivated by the challenge of work that it takes to reach the aspired level of performance, and thus become “psyched up”. Furthermore, star “prestige” name on the work of co-authors may increase their chances to attain funding for journals in the future. Therefore, they can be motivated to take in star shared knowledge due to the promotional opportunities that the prestige name of a star offers. This can contribute to the main effect in two ways. On the one hand, it can make human capital spillovers more effective. On the other hand, co-authors may produce more work due to the exploitation of the opportunities that the promotion has given post working with a star. Lastly, as the presence of a star has established a high reference point, to aspiration to get there should trigger co-author motivation for sense of achievement.
I propose that co-authors will have higher output of innovative ideas not only due to unique knowledge inputs, knowledge recombination, and close collaboration but also due to star motivation to share novel knowledge with co-authors and co-author’s motivation to intake star-shared knowledge.
Hypothesis 1: Post working with a star has a positive effect on the co-author’s output of innovative ideas.
2.5 Co-authors and Variance Between Spillover Effects
Besides the newly proposed reasoning of knowledge sharing motivation, it is evident across multiple past studies that there are human capital spillovers when stars work with co-authors. However, despite the empirical evidence, what still remains pressing is the fact that there exists variance of the human capital spillover effect. Furukawa & Goto (2006) found that star scientists have positive effect on co-author patent output. Azoulay (2010) discovered that when a star scientist dies, their co-author paper publishing rate decreases by 5% to 8%. Oettl (2012) showed a different effect. When a star died, not the quantity but only the quality of their author paper output decreased. It has also been evident that the decline is experienced in co-author’s citation-weighted patents and lifetime earnings when a star unexpectedly dies (Jaravel, Petkova, & Bell, 2016b). What is surprising is that other studies do not find similar effects. Research on historical accounts found that when Jewish scientists emigrated to the United States, they did not affect the incumbent US scientists in the universities where the emigrated stars resided in. Instead, they attracted and positively influenced the patenting output of other scientists who travelled across the country to work with the new arrivals (Moser et al., 2014). Additionally, another historical study also did not find any effects in co-worker productivity change when the Nazi Germany regime dismissed a group of star scientists from their universities (Waldinger, 2012b). It is understood what is happening when stars work with
co-authors. However, it remains unknown when the effects are happening. Varied research results suggest that there are other phenomena at play which determine if spillover effects happen or not.
2.6 Co-authors and Influence by Social Comparisons
One of potential explanations of the main effect variance could be social context mechanisms that are known to be present in academic environments. There is evidence that scientists “race” with each other or try to “copy” the activities of more successful scientists in their social network. Research on social context within academic departments shows that depending on what scientist’s peers are doing, the focal scientist will try to “copy” his peers. Co-authors adjust their behaviour and engage in applied research activities due to the social context in which they operate in (Bercovitz & Feldman, 2007; Louis, Blumenthal, Gluck, & Stoto, 1989; Stuart & Ding, 2006; Tartari et al., 2014). This mechanism is regarded as imprinting, which is the process during a specific time period wherein an individual develops characteristics that reflect features of the environment (Aschhoff & Grimpe, 2014; Marquis & Tilcsik, 2013). Prior studies have found that academic individuals are more likely to become entrepreneurial (join boards or commercial firms) if they have peers in their university departments who have engaged in similar activities in the past. Effect becomes even stronger if these peers are perceived as prestigious (Stuart & Ding, 2006). These studies have focused on “localized” peer effects, which are referring to the people in a scientific department. In this study, I examine “personal” peer effects which are influences from person-to-person collaborations (Aschhoff & Grimpe, 2014). As imprinting effects come from the environment, in this case the environment is the prestige-status star, and the time period is the period of the
Research on social comparisons explains the phenomenon in more depth. During the process of imprinting, individuals compare themselves with their peers and establish a perceived gap between their status and the status of a peer (Tartari et al., 2014). Social comparisons become especially apparent when a scientist is working with a high performing scientist, as the perceived gap is even bigger. I previously discussed the study from Flynn & Amanatullah (2012) which states that individuals set higher goals for themselves because they see high-status performers as “anchors”. I expand on this phenomenon, explaining that the variance that the co-authors experience may happen due to the differences in the gaps that stem from social comparison with a star.
Individuals derive their self-worth by comparing themselves with peers in similar stage in their careers. Social comparison theory presumes that individuals have natural inclination to “self-evaluate” themselves. They do so by comparing their achievements and abilities to the ones of their peers. Peers of similar rank and similar attributes become reference points in the individual’s social context. An individual refers to the reference points and tries to replicate their “successful” behaviour to reduce the perceived difference between their achievements and the ones of their peers. However, objective equilibrium is difficult to assess and thus, to ever close the perceived gap is impossible. As a result, continuous competition happens. Peer influence and competitive behaviour are the drivers under social comparison. I assume that co-authors and stars are of similar rank and attributes. If a star would perceive co-author as incompetent, they would not be working in the same team. However, a co-author still remains of a lower status than the star. Moreover, it cannot be that all co-authors will have equal differences between their status and the status of a star scientist. Some people simply have had more career success than others. Thus, some co-authors are bound to experience bigger “gaps” than other co-authors. As a result, some co-authors are likely to be more productive than others. But what could be a good indicator to the size of these gaps?
2.7 Co-author’s Age as a Proxy for Social Comparisons
Research in academic entrepreneurship has studied how scientist’s personal characteristics, such as age, influence his or her industry engagement “copying” behaviour (Aschhoff & Grimpe, 2014; Bercovitz & Feldman, 2007; Colyvas & Powell, 2007; Haeussler & Colyvas, 2011; Louis et al., 1989; Stuart & Ding, 2006). Industry engagement is referred to as applied science activities such as conducting research within companies, partnership and sponsorship with commercial organizations, employment by profit or commercially focused organizations, and number of patent publications. As this research is looking into the generation of innovative ideas, I only use the patent dimension of industry engagement. It is also important to note that these findings are drawn from studying how industry engagement affects the industry engagement of others. As previously discussed, stars tend to take scientific roles rather than applied science roles and therefore are likely to be more science engage than industry engaged. If co-authors would imprint or copy star behaviour, they would produce academic papers. I assume that co-authors are unlikely to do so and rather take applied science opportunities due to their mid-level social status. Therefore, I assume that co-authors “copy” star behavior in the way that is feasible for them.
Regarding the “copying” behaviour affected by age, one particular research has examined these effects in person-to-person collaborations (Aschhoff & Grimpe, 2014). This study uncovers an interesting moderating effect. The lesser the career-age of the scientist who collaborates with an industry-engaging peer, the more likely he or she is to imitate the behaviour of the peer and engage in applied science activities as well. In contrast, scientists who have been for decades in academia had no influence on them at all. Others who studied
patent publishing, but finds no effect. Survey in research university among life science faculty members finds that gender and age are weak predictors towards any scientifically commercial activity (Louis et al., 1989).
An insightful observation can be made when taking these studies into perspective. They examine the influence from departmental peer effects and how age and experience moderates it, except the one from Louis et al. (1989) who take academic’s age as an independent variable. It is evident that age on its own is not a strong predictor to production of innovative ideas. However, if an individual’s age is a moderator to the peer effect, age becomes an important variable. Thus, to answer the earlier stated question to what could be a good indicator to the differences in the gap sizes resulting from social comparisons, I propose that age could act as a proxy for the gap size resulting from social comparisons. Co-author who is of young age has not had the same amount of time in his career as an older co-author who has been working at their career for years. Therefore, when a young co-author compares himself with a star, he may experience larger gap than an older co-author. As a result, the younger co-author could be more motivated to produce innovative ideas, codify them into patents, and lessen the gap between him or her and the star. For older co-authors, this would not be the case to the same extent. Human capital spillover effects are very likely to happen, but due to the lesser gap between them and the stars, they may experience the spillovers to a lesser extent and therefore be less motivated to be productive post working with a star. This proposition could potentially explain a part of the existing variance between all co-authors who have worked with a star.
Besides age acting as proxy for social comparison gaps, prior research has indicated that personal age also correlates with individual productivity (Skirbekk, 2004). It has been observed that scientific productivity decreases with age (Bonaccorsi & Daraio, 2003; Cole, 1979; Costas, van Leeuwen, & Bordons, 2010). Skirbekk (2004) argues that older individuals experience loss in cognitive ability to process and learn new information and are more
productive in familiar, past-experience based tasks. Whereas younger individuals are better performing at tasks which require to create novelty. In addition, Bonaccorsi & Daraio (2003) propose that the age difference is also associated with differences in income. Older scientists do not engage in applied research activities as much as younger scientists. Scientists of old age are more established in their income. They are likely to focus on distribution and teaching of knowledge, whereas younger scientists tend focus on exploiting their knowledge in commercial activity to attain income.
Hypothesis 2: Co-author’s age has a moderating effect on his/her generation of innovative ideas post working with a star.
3. Methodology
This chapter elaborates on the methods used to conduct the research. I provide an overview of the research population and the research sample. I also describe the data collection process for independent, dependent, moderating, and control variables.
3.1 Population
In order to know the co-authors of stars, stars themselves must be identified first. The Atomic Heritage Foundation (AHF) provides access to historical archives, which consist of scientists who worked in the Manhattan Project. Manhattan Project was a United States led research and development programme during the 2nd World War from 1939 till 1945 with a goal to produce an atomic bomb. As a massive scale project which was advancing nuclear science, it required the best performing scientists who would have the necessary capabilities to work on it. Scientists were selected from a list of top candidates and contacted personally, being called out for war duty. Candidates themselves were unaware of the selection and had no influence in the selection process. This is important as it allows us to observe peer effects from any interfering factors. For example, when studying the effects of collaboration, stars can be selected due to their collaborative abilities and not due to their quality of work. This would exclude a whole population of stars who do not cooperate well with others and therefore the results from star spillover effects would be biased.
Furthermore, the AHF data archives have classified 539 people as scientists. However, not all data on these individuals is accessible through non-paying means. 359 Manhattan Project participants are identifiable through free access on Microsoft Academic Search. This study accesses the data of the 359 participants in a pre-made database. By knowing the identities of stars, their respective patent co-authors can be obtained in patent databases. I will discuss the method of co-author data collection in the section of data collection process.
Manhattan Project participants and their patenting co-authors is a fitting population for this study. Firstly, the candidate selection criteria are a good match with the star classification of prior research. Stars are categorised as individuals of disproportionate productivity in comparison to their peers, and disproportionate visibility and attention within their industry or field (Groysberg et al., 2008). Manhattan Project participants would not be able on the candidate list if they could not be able to generate visibility in their fields. Their visibility most likely comes from the high-quality output of their work. Secondly, the project avoids the selection effect. The selection is not influenced by other capabilities of a star, such as the ability to work well with others. Thus, selection effect is not exogenous to the quality criteria, ensuring high probability that stars are actually stars. It is still important to acknowledge that selection effects cannot be eliminated completely. Thirdly, the population fits the star knowledge sharing motivation factor “perceiving knowledge as of high value” i.e. when knowledge is novel (Ipe, 2003). Manhattan Project conducted top-tier research and development in nuclear science. The made discoveries were kept in high secrecy after the end of the project. Therefore, the knowledge is novel as there are no other scientists than the stars who were holding the knowledge accumulated from the project.
3.2 Sample
From the 359 Manhattan Project participants, 30 stars were selected through systematic sampling. Starting from the top of an alphabetically arranged (A to Z) list, every 4th star was selected. From the sample of 30 stars, all their respective patent co-authors can now be identified by collecting star patents from the Google Patents database. Table 1 showcases all the sampled stars and their patenting co-authors.
searched in the Google Patent database and their patent data was downloaded as a .csv data file (option provided by Google). Google Patent database covers 87 million patents, uses other patent databases (USPTO, EPO, WIPO, DPMA, CIPO, and SIPO) and provides full patent scans from 17 patent offices. Therefore, it is an efficient and reliable source of patent data.
As data was collected, occasional same surname problems appeared and data cleaning was employed. To ensure data validity, each star’s name was checked in the AHF online database. Once their approximate expertise was known, the stars were double checked by seeing if their expertise matches the one in the patent. If the AHF provided a date of birth and a date of death of a star, patent registration date could be checked and validated if it falls in the range of a star’s lifetime.
From the collected data, a database was constructed. 12 out of 30 stars had a published a patent at least once in their lifetime. In the database, it is observable how many registered patents the star has and with whom the star co-authored the patents with. All of the co-author names were separated from the ones of stars, and listed in a table (Table 1). To collect co-author patenting data, the same data collection and cleaning procedure was applied as the one to stars. Moreover, each co-author was also cross-referenced in the AHF database to ensure that he or she is not actually a star. There were total 42 co-authors identified from which 9 were star scientists in the Manhattan Project (therefore, they are excluded from the hypothesis testing). The number patents per co-author ranged from 0 to 6. Just as stars, co-authors were searched in the Google Patents database and their patent output data was collected. Lastly, once the patent data of all stars and co-authors were accumulated into one database, the data is cleaned by checking for patent doubles and for correct name spelling of stars and co-authors across all patent cases.
The co-authors of stars also had co-authors (named co-co-authors) of their patents. These individuals were also incorporated in the database and used as a baseline group (group that never has worked with a star) to study the human capital spillover effects from a star.
Finally, the patent dates were rounded to whole years and the database was transformed into longitudinal database. As this study examines time effects, patents without date information were dropped. The database holds 518 unique patents. Instead of each row being a patent, each row is a single-year observation of a particular author. For each author, the years range from 1882 to 2009. As star patent output is not measured in this study, it is dropped out from the analysis. In total, there are 206 unique authors, consisting of co-authors and co-co-authors.
Table 1
Overview of Stars and Co-authors
Star Patent Co-author Star
1. Willis A Adcock Yes 1.1 Raymond C Sangster
1.2 Frank L Skaggs 1.3 William C Lake 1.4 Morton E Jones No No No No 2. Harold M Agnew No 3. Donald E Ames No 4. David S Anthony No 5. Jane R Arnold No
6. Dale F Babcock Yes 6.1 Charles W J Wende
6.2 Arthur W Larchar 6.3 Stnart J Bugbee 6.4 Victor F Hanson 6.5 Crawford H Greenewalt 6.6 Wortington Hood 6.7 John S Neil No No No No Yes Yes No 7. Berry W Bailey 8. Claire C Balke 9. Benjamin Benderson 10. Francis Birch 11. Thomas W Bloss 12. Aage Niels Bohr 13. Norris Bradbury 14. Leo Brewer 15. George Brooks 16. Solomon Bunshaft 17. Morton Camac 18. Chamberlain Owen
19. Marion Edward Cieslicki
20. Burgess F Collins 21. Edward U Condon 22. Elwin H Covey 23. Edward C Creutz 24. Winston Dabney 25. John W De Wire 26. David M Dennison 27. Lee A DuBridge 28. Freeman J Dyson 29. Spofford G English 30. Frank Farrar Yes Yes Yes No No No Yes No No Yes Yes Yes No Yes No Yes No No No No No Yes No 7.1 Lyle J Gordon 8.1 Clarence W Balke 8.2 Misegades Keith 8.3 William S Graff 9.1 Eisinger Joseph 9.2 Rubin Kenneth 14.1 George R Bird 14.2 Jr Willes E Grey 17.1 Fritz Bien
17.2 Michael Elliot Gersh 17.3 John H Caulfield 18.1 Clyde E Wiegland 18.2 Emilio G Segre 18.3 Carson D Jeffries 19.1 Eugene Ludwick Dunn 19.2 James E Wilson 19.3 Joseph Burton Moore 19.4 Benny J Nelson 21.1 Gereld L Tawney 21.2 Willard A Derr 23.1 Edward H Zinn 23.2 Enrico Fermi 23.3 William A Mcadams 23.4 Martyn H Foss 23.5 Walter B Shank 23.6 David H Gurinsky 23.7 Szilard Leo 23.8 Eugene P Wigner 23.9 Leo A Ohlinger 23.10 Gale J Young 23.11 Alvin M Weinberg 29.1 Jack K East No No No No No No No No No No No No No No No No No No No No Yes Yes No No No No Yes Yes Yes Yes Yes No
3.4 Measurements
3.3.1 Dependent Variable - Generation of Innovative Ideas
My argument was that co-authors are more likely to take the role of knowledge exploitation and thus engage in applied science activities or commercial activities. These activities require the generation of innovative ideas. Academic entrepreneurship literature measures these activities by “commercial activity index”, which consists of 3 variables: involvement of consulting companies, patenting publishing, and firm founding (Haeussler & Colyvas, 2011). However, this data is usually acquired by the means of survey. This study looks at historical data from the stars of AHF, where the co-authors of these stars lived in the 20th century. There is insufficient data to cover all variables in the commercial activity index. Only patent information is accessible, accurate, and reliably kept in Google Patent database. For this reason, this study measures generation of innovative ideas by number of patents published.
Patenting activity on its own provides a reliable measure for generation of innovative ideas (Acs, Anselin, & Varga, 2002). Patent publishing shows the closing process of innovative idea generation. From the start of acquiring knowledge and then recombining it, patent acts as the final form of knowledge codification and documentation. Therefore, it is a strong and accurate measure of innovative idea generation.
What is important to take in account is the fact that patenting is a rare event (Kinney, Krebbers, & Vollmer, 2004). Scientists in general patent significantly less than they publish papers. This is bound to create normality issues in the data, as there is likely to be substantially
3.3.2 Independent Variable – Co-author post star
To study the effects of post working with a star, each co-author’s timeline in the panel data is divided in two parts. The time before they have worked with a star versus the time after. This is also known as the time varying post variable. The point of separation is the first year after publishing a patent together with a star for the first time. It is a dichotomous variable taking two values (0;1). Having the two separate timelines allows to indicate the number of patents published in each timeline. In turns, this allows to observe if the patent output is larger after working with a star than it was before. If there are any significant changes, working with a star would partly explain them.
3.3.3 Moderating Variable – Academic age
Co-author’s age acts as a proxy for social comparison gap sizes. To measure age very accurately, birth date of each co-author would be needed. However, such data is only available for very few co-authors. The rest of the co-authors would have missing values. Therefore, I follow the operationalisation methods of prior research studying the individual effects of scientist age (Bonaccorsi & Daraio, 2003). By taking the very first year when co-author publishes a patent, the age counting initiates. For the year a patent is published, the age is 1. Every following year adds an additional value of 1 to the career age. The variable moderates the main effect and would potentially explain that the decrease in patenting output is partly due to the increased age of the co-author. In turn, this would explain why some co-authors are more productive than others after working with a star.
One could argue that there are limitations to such methodology. In a case where a co-author’s actual age is 60 but he or she only publishes a patent now, co-co-author’s age would be indicated as 1. In contrast, in a year wherein another co-author publishes a patent while his actual age is 30, would also have his or her age indicated as 1. Thus, measuring by academic age does not accurately reflect scientist actual age. However, this study proposes that social
comparison gap is a result from perceived difference in career achievements and that young co-authors would be motivated to close the gap. Career achievement differences do no stem from the actual age but from how many career years a co-author has had to make achievements. 3.3.4 Control Variables
1. Collaboration continuity – When two or more people work together across long periods of time, they build understanding of each other, knowing how to communicate and what working methods are the most effective for each individual. They create systems of phrasing, symbols, and behaviours to work more efficiently. As a result, they develop in-group tacit knowledge (Burt, Kilduff, & Tasselli, 2013). Hence, these individuals are likely to become more productive over time when working together. In the context of this study, finishing one patent allows to be more productive working on the next patent. For this reason, it is important to control for collaboration continuity, as it can explain the variance in increased patent output by co-authors. Co-author patenting increase might not be due to the human capital spillovers but due to continued collaboration. When two or more individuals have authored on two or more different patents, the variable value takes 1. Otherwise it takes a 0.
The variable list also includes team continuity variable, which indicates when the same exact group works on another patent. This is a stricter control and would not capture the effects when only two people work together again and the third one is a newcomer. Two individuals can still carry on their tacit knowledge to their next patenting activity. Therefore, team continuity variable is excluded.
2. Post Manhattan Project – All scientists patented more after the Manhattan Project than they did before it. Most likely this is due to the major advancements made in nuclear
4. Results
I first start by describing the descriptive statistics and correlation analysis. Then, I discuss the results of linear regression analysis for Hypothesis 1 and Hypothesis 2, followed by robustness checks.
4.1 Descriptive Statistics
Table 2 marks the overall frequencies of the independent variable and frequencies between groups before working with a star and after working with a star. The sample consists of 206 scientists who are in two groups. First group are the 206 (100%) scientists who have not yet worked with a star. The second group are the 37 (18%) scientists who later in their lives have worked with a star. The remaining 169 scientists never worked with a star and remain as a baseline for hypothesis testing.
The descriptive statistics of mean, minimum and maximum values, standard deviations, and bivariate correlations are displayed in Table 3. Between years 1882 and 2009, there are 23,078 author-year observations. Observations start from year 1882 only for 3 scientists. For the rest, the observation period is from year 1898 to year 2009. The patent output ranges from 0 to 8 patents (M = .029, SD = .245). The mean is very low due to the data with large amount of 0 relative to the values of 1. Therefore, Breusch-Pagan/Cook-Weisberg test is used, checking for heteroscedasticity. It is found that there is exists significant skewness in the data. The patent output variability is unequal across the range of values of the predictor variable. This indicates that significance and effect coefficient in regression results can be heavily inflated. Thus, on top of testing hypothesis, robustness checks will be employed to observe if the regression effects will remain significant. Furthermore, academic age in the sample ranges from 0 to 128 (M = 12.06, SD = .245). The mean is small if compared with the maximum age. This is due to the large amount of 0 values.
In addition, we can observe if there is shared variance between any two variables by looking at the Pearson correlation matrix in Table 3. Generation of innovative ideas is positively correlated with collaboration continuity (r = .52, p < .01), team continuity (r = .46, p < .01), post MP (r = .03, p < .01), and co-author post star (r = .14, p < .01). The relations with the first two variables is two be expected. Whenever a scientist publishes a patent, it is often in collaboration with others. Team continuity is simply a stricter variable of the same variable (measuring the same exact team across patents). Correlation with post MP indicates that those co-authors who collaborated with stars, may experience the unlimited resource effects which stars attained during their participation. Moreover, we can also see that the independent variable co-author post star correlates with every variable in the list. Highest correlations are with academic age (r = .39, p < .01) and post MP (r = .25, p < .01). This is to be expected for age, as in a year in which a co-author works with a star, academic age has either value of 1 or higher. Correlation with post MP could be due to stars returning from Manhattan Project and collaborating with non-stars in their respective organisations.
Due to the many correlations, Variation Inflation Factor (VIF) is used to test for multicollinearity. The test returns values ranging from 1 to 3.19, meaning there is no multicollinearity among the variables. Nevertheless, collaboration continuity and post MP are used as controls. Team expertise (r = -.02, p < .01) is not used as a control variable as it only accounts for a very small fraction of the shared variance.
4.2 Regression Analysis
The next steps in data analysis will focus on hypothesis testing, which will be done by means of hierarchical multiple regression while controlling for collaboration continuity and
However, before testing the hypothesis, Model 1 is constructed only with controls collaboration continuity and post PM, which predict generation of innovative ideas (Table 4). The tested model was statistically significant F (3, 205) = 97.82, p < .01, and explained 27.2% of variance in the co-author’s output of innovative ideas. The control variable post MP was not statistically significant (b = .002, p = .67). While collaboration continuity appears to be a significant predictor (b = 1.40, p < .01) of generation of innovative ideas. This predictor explains surprisingly large variance. It could be the case that when scientists continuously work together, they learn how to work efficiently and be more productive, allowing them to produce more patents over time. It can be that collaborating scientists agree to stick together for a longer time and output patents in bursts. Controlling for time gaps between patents and seeing if the same scientists work subsequently after one patent and another would explain this effect better. This could also be explained in more detail by having controlling for a variable which measures if these continuous collaborations are with stars or non-stars.
4.2.1 Hypothesis 1
The first hypothesis states that post working with a star has a positive effect on the co-author’s output of innovative ideas. Model 2 is constructed by three predictors: co-author post star, post PM, and collaboration continuity, which predict generation of innovative ideas (Table 4). This model is used to determine if, after working with a star, the innovation output of co-authors would be larger than before working with a star i.e. if there are human capital spillover effects. The tested model was statistically significant F (3, 205) = 120.33, p < .01, and explained additional 1.4% of variance in the co-author’s output of innovative ideas. In this model, only the control variable post MP was not statistically significant (b = -.01, p = .08). While collaboration continuity appears to be the most significant predictor (b = 1.38, p < .01) of innovative idea output. In line with Hypothesis 1, working with a star is a significant predictor of co-author’s output of innovative ideas (b = 0.1, p < .01). In other words, co-authors
who work with a star produce 0.1 more patents than co-authors who never have worked with a star. Scientists who work with a star are slightly more productive of the generation of innovative ideas. Therefore, based on the outcome of Model 2, Hypothesis 1 can be confirmed. Although, the effect of working with a star is not as large as expected. Potential cause of this could be in the methodology. Patent publishing is a rare event and measuring by productivity rare occasions may not capture the actual spillover effects. Enrichening the data with the number of academic paper published could improve accuracy.
4.2.2 Hypothesis 2
The second hypothesis states that co-author’s age has a moderating effect on his/her generation of innovative ideas post working with a star. To test the moderating effect of age, an interaction term is created. The interaction term represents the moderating effect. It was coded as interaction = co-author post star * academic age. Building upon Model 2, Model 3 was created by removing variable co-author post star and adding the interaction term to the predictors list (Table 4). The model proved to be significant F (4, 205) = 91.78, p < .01, but explained only additional 1.1% of variance in output of innovative ideas. The interaction had a significant negative relationship with output of innovative ideas (b = -.003, p < .01). Meaning that when co-authors who worked with stars got older were less productive in generating innovative ideas. Therefore, Hypothesis 2 is confirmed. However, the loss in productivity due to the aging effects seems to be very minor. This could be due to the fact that there is skewness in the academic age variable. As it only shows values above 0 when there is a patent published, there are significantly more 0’s than other values.
results can be derived. I address the skewness issues in the next section by using more rigorous regressions. The outcomes of Models 2 and 3 would then be compared to the outcomes of the robustness check regressions.
4.3 Robustness
Three robustness checks were employed, which were regression using logged dependant variable, probit model regression, and comparison of means.
Firstly, 0 values were dropped for the variable academic age. This resulted in 13,682 dropped observations. Secondly, for the dependent variable, a log was created by using formula generation of innovative ideas+999, followed by log (generation of innovative ideas). Robust clustered regressions were again conducted for Model 2 and Model 3. Model 2 remained significant F (3, 205) = 119.05, p < .01, and explained 30.4% of the variance, which is by 1.8% more than the initial regression without robustness check. The predictor co-author post star also remained significant (b = .01, p < .01), but the effect size is clearly reduced. We can also observe that the standard deviation of generation of innovative ideas is greatly reduced from .245 to .0004. We can infer that the skewness in data had inflated the effect size. If the standard deviation was driving the significate results, it is no longer the case. Model 2 is still significant and its significance was not due to chance. Furthermore, Model 3 also remained significant F (4, 205) = 93.81, p < .01, and explained 31.3% of the variance, which is by 1.6% more than in the non-robustness regression. The interaction term did not lose significance but the effect size was substantially reduced (b = -3.37e-06, p < .05). It seems that the b coefficient was massively inflated due to the skewness in the data. Nevertheless, Model 3 also passes the robustness check and endures to be significant.
Additionally, another robustness check was conducted by using the probit model regression. The model is a binary classification model, which estimates the probability to which an observation will fall in one of the two values of a dependent variable. Therefore, the
generation of innovative ideas variable was transformed into a dichotomous variable, where value 0 stands for no patent published and value 1 stands for patent published. Collaboration continuity control was dropped as it takes a value 1 whenever dependent variable is 1. The probit regression was run with the dropped academic age 0 values for Model 2 and Model 3. The probit model from Model 2 was a significant fit to the data (Wald chi2 = 133.61, p < .01). Co-author post star was positively significant (b = .48, p < .01) predictor to generation of innovative ideas. Unlike OLS regression model, probit model b values are interpreted as more likely probability if b > 0, or less likely probability if b < 0. Whereas p < .05 indicates significance at the 5% level. Therefore, we can interpret from the results that co-authors who have worked with a star, in comparison to scientists who have not, are more likely to publish patents. Regarding Model 3, the probit model was also significant (Wald chi2 = 140.06, p < .01). The interaction term was negatively significant to generation of innovative ideas (b = -.02, p < .05). Hence, co-authors who have worked with a star and are of older age are less likely to publish patents. Probit model robustness check addressed the issue of dependant variable skewness and confirmed that Hypothesis 1 and Hypothesis 2 results remain significant under more rigorous test.
Lastly, generation of innovative ideas means were compared by checking the mean of patents for authors who have not worked with a star versus the mean of patents for authors who have worked with a star. The mean of patenting for star co-authors is higher (M = .14, SD = .60) than it is for authors who have not worked with a star (M = .05, SD = .36). Thus, authors who have worked with a star produce more patents on average than authors who have not worked with a star. As there exists a difference in these means, it can be surely interpreted that
Table 2
Summary of independent variable co-author post star
Co-author Freq. Overall Percent Overall Freq. Between Percent Between Within Percent
Before working with a star 21102 91.44 206 100.00 91.44
Post working with a star 1976 8.56 37 17.96 47.64
Table 3
Descriptive statistics and Pearson correlation matrix
N Mean S.D. Min Max 1. 2. 3. 4. 5. 6. 7. 8.
1. Generation of Innovative Ideas 23,078 .0294 .245 0 8 1.00 2. Academic Age 23,078 12.060 19.876 0 128 -.01 1.00 3. Collaborator Continuity 23,078 .008 .091 0 1 .52** -.01 1.00 4. Team Continuity 23,078 .006 .076 0 1 .46** -.01* .83** 1.00 5. Post MP 23,078 .553 .497 0 1 .03** .50** .04** .02** 1.00 6. Team Expertise 23,078 .117 .298 0 1 -.01 -.03** -.01 -.01 -.00 1.00
7. Co-author Post Star 23,078 .087 .280 0 1 .14** .39** .07** .05** .25** -.02** 1.00 8. Co-author Post Star *
Academic Age
23,078 3.178 12.525 0 127 .40** .52** .02** .01 .22** -.01 .82** 1.00 Note. N = 23.078 observations, n = 206 observations
Table 4
Results of the regressions for the predictors of generation of innovative ideas
Variables Model 1 Model 2 Model 3
Collaborator Continuity 1.40** (.100) 1.38** (.095) 1.36** (.089)
Post MP .002 (.004) -.012 (.007) -.010 (.006)
Co-author Post Star .100** (.029) .233** (.081)
Co-author Post Star * Academic Age -.004* (.002)
Model F 97.82** 120.33** 91.78**
R2 .272 .284 .295
N 23,078 23,078 23,078
Note. Standard errors are in parentheses.
**Correlation is significant at the .01 level (2-tailed) *Correlation is significant at the .05 level (2-tailed)
5. Discussion
There is a wide agreement across studies that stars produce human capital spillover effects on their co-authors and make them more productive towards the generation of innovative ideas. On top of the conventional argument, this study also proposed that the effects will happen when stars are motivated to share novel knowledge. Additionally, despite the empirical evidence of the main effect, past studies show that the spillover effects are unevenly distributed. Not all co-authors equally benefitted by working with stars. This study attempted to address the issue by arguing that co-authors experience differently sized social comparison gaps when comparing their achievements to stars, and that co-author response to the gaps could partly explain the variance in productivity. I proposed that the age of co-authors could act as proxy for the gap sizes. I hypothesized that younger co-authors have less career achievements, experience larger gaps than older co-authors do, and therefore generate and codify innovative ideas to lessen the perceived gap between them and the stars. Older co-authors would experience smaller gaps and therefore would generate less innovative ideas than younger co-authors.
Using a dataset of patent information of co-authors who worked with Manhattan Project participants, it was found that post working with a star has a positive significant relationship on co-author generation of innovative ideas. Initial hypothesis testing showed that post collaboration co-authors published 0.1 more patents for every patent that individuals who never worked with a star did. However, it was inferred that skewness in the data inflated the effect size. After robustness checks, the hypothesis remained significant, yet the effect size was heavily reduced. The findings are consistent with the argument of other studies, claiming that
other studies where co-author productivity loss after death of a star ranged from 5% to 8% (Oettl, 2012). We cannot be sure of the actual effect size due to the skewness, but the 0.1 value in the results was heavily inflated. Still, patenting is a rare event and even the post-robustness check 0.01 value is meaningful for patenting. Possible explanation to why effect size is smaller than prior studies could be that patenting is not the only way how co-authors exploit knowledge. Prior studies indicate other innovative idea generation activities in which scientists engage. Academic entrepreneurship literature has indicated these activities as consultation for firms and firm founding (Haeussler & Colyvas, 2011). It could also be that the argument stated in the literature review that co-authors tend to focus on applied sciences is invalid. Some studies show that science and technology is closely interlinked, and that a scientists refer to papers when making patents (Grupp & Schmoch, 1992; Narin & Noma, 1985). In various fields of science, papers are published first to build knowledge and later come patents that refer to the knowledge (Carpenter, Cooper, & Narin, 1980). The fact that the Manhattan Project participants were experts in a relatively new field of nuclear science could mean that, instead of patents, co-authors took the knowledge of stars and published papers. There could have been little knowledge available to build patents from.
Furthermore, the moderating effect of co-author academic age was examined on the relation between post working with a star and generation of innovative ideas. The results indicate that co-author’s age has a significantly negative effect, even after robustness checks. Older co-authors do indeed publish less patents than younger co-authors do. This is in line with research saying that the lesser the career age of a scientist, the more likely he or she is to engage in a “copying” behaviour the peers that the scientist worked with (Aschhoff & Grimpe, 2014; Bercovitz & Feldman, 2007). In the context of this research, it is co-authors “copying” the career-achieving behaviour of stars. Similar to the main effect findings, moderating effect size was also smaller than expected. The model explained only marginal 1.1% of variance in
generation of innovative ideas. Nevertheless, this finding does contribute to the theoretical gap and partly answers the question when do co-authors experience spillover effects to different extents. I claimed that age acts as a proxy to the size of the gaps resulting from social comparisons with stars. Whether this is true or not, we cannot be sure. But evidence shows that there is indeed something going when older co-authors work with stars.
Besides hypothesis testing, what was unexpected was that collaboration continuity accounted for a large portion of the variance. This is interesting as the observation suggests that collaborating with the same individuals is more effective than working with a star. I argued that close and frequent team collaboration is an excellent condition for knowledge transfers (Groysberg et al., 2008; Schiffauerova & Beaudry, 2011). Therefore, continuing working with the same individuals may be more effective for generation of innovative ideas than just collaborating with stars. However, it is important to acknowledge that it was not tested whether the same collaborations are done with stars or non-stars. More accurate observations could be made when each of the two collaboration types are controlled for separately.
In sum, we can conclude that co-authors respond well to the presence of stars, especially when they are of younger age. They become more productive towards the generation of innovative ideas. Other studies who also studied this peer effect, did so by measuring co-author productivity decrease when a star scientist passes away or leaves geographical proximity. This study went against the conventional method and added evidence to the literature by showing how productivity increases (rather than decreases) post working with a star. This adds an insight of the productivity change when the star is still around. Prior literature also indicated that spillovers happen when stars are of collaborative behaviour and relationship-building in
