Management of Science, Serendipity, and Research Performance: Evidence from a Survey of Scientists in Japan and the U.S. Kota Murayama a,* , Makoto Nirei b , Hiroshi Shimizu b a Department of Economics, Northwestern University, 2001 Sheridan Road, Evanston, IL 60208, United States b Institute of Innovation Research, Hitotsubashi University, 2-1 Naka, Kunitachi, Tokyo 186-8603, Japan Abstract In science, research teams are increasing in size, which suggests that science is becoming more organisational. This paper aims to empirically investigate the effects of the division of labour in management and science on serendip- ity, which has been considered one of the great factors in science. Specifi- cally, in examining the survey of scientists conducted in Japan and the U.S., this paper treats the following questions: Does pursuing serendipity really bring about better scientific outcomes? How does the division of labour in science influence serendipity and publication productivity? The empir- ical results suggest that serendipity actually brings about better research quality on average. It also finds that if the managerial role is played by a leading scientist in the team, this is positively associated with the quality of the paper through allowing researchers to pursue serendipitous findings. In contrast, if the managerial role and leading research role are played by different members, this has a positive association with the number of papers published, as the project size becomes larger. These results indicate there is * Corresponding author. E-mail: [email protected]Preprint submitted to Research Policy October 11, 2014
49
Embed
Management of Science, Serendipity, and Research ...nirei.iir.hit-u.ac.jp/papers/MoS.pdf · First, serendipity has a positive association with citation of papers. This suggests that
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Management of Science, Serendipity, and ResearchPerformance: Evidence from a Survey of Scientists in
Japan and the U.S.
Kota Murayamaa,∗, Makoto Nireib, Hiroshi Shimizub
aDepartment of Economics, Northwestern University, 2001 Sheridan Road, Evanston, IL60208, United States
bInstitute of Innovation Research, Hitotsubashi University, 2-1 Naka, Kunitachi, Tokyo186-8603, Japan
Abstract
In science, research teams are increasing in size, which suggests that science
is becoming more organisational. This paper aims to empirically investigate
the effects of the division of labour in management and science on serendip-
ity, which has been considered one of the great factors in science. Specifi-
cally, in examining the survey of scientists conducted in Japan and the U.S.,
this paper treats the following questions: Does pursuing serendipity really
bring about better scientific outcomes? How does the division of labour
in science influence serendipity and publication productivity? The empir-
ical results suggest that serendipity actually brings about better research
quality on average. It also finds that if the managerial role is played by a
leading scientist in the team, this is positively associated with the quality
of the paper through allowing researchers to pursue serendipitous findings.
In contrast, if the managerial role and leading research role are played by
different members, this has a positive association with the number of papers
published, as the project size becomes larger. These results indicate there is
Medicine & Psychiatry/Psychology, Agricultural Sciences & Plant & Animal
Sciences, Basic Life Sciences, and Social Sciences.4
3.4. Estimation Models
We employ the following estimation models to examine our three hy-
potheses. The H-1 model investigates the relation between serendipity and
research quality. In estimating the effect of serendipity on quality, we pay
particular attention to the possibility that serendipity has an endogenous
relation to the quality. As we reviewed above and will discuss in the fol-
lowing section, a researcher pursues publication of a scientific finding only
when the researcher believes that it is worth pushing forward, while the re-
searcher’s ex ante evaluation of the finding may differ depending on whether
the finding emerged from serendipity. If this is the case, the coefficient of an
ordinary least squares (OLS) regression may overestimate or underestimate
the effect of serendipity. To circumvent the endogeneity bias, we use a Two-
Stage Least Squares (2SLS) regression, in which we first regress serendipity
on the instrumental variables, and then the number of citations is regressed
on the predicted serendipity of the first stage. In this way, the coefficient
in the second stage captures the effect on the number of citations of the
exogenous variation in serendipity.
The H-2 model asks whether the integration of core-scientists with man-
agement affects serendipity. Since serendipity is a binary variable in our
dataset, we conduct a probit regression. An important issue with probit
4Social Sciences may be a fairly broad field compared to other fields. However, sinceabout 95% of the respondents are natural scientists, this makes no significant difference.
19
regressions is their fragility to any heteroskedasticity in the error terms.
Hence, we test whether our results are robust to misspecification of the
error term, and claim that our hypothesis still holds.
For the H-3 model, which examines the relation between management–
research separation and research productivity, we use a Negative Binomial
(NB) regression under the assumption that the variance of the dependent
variable takes a quadratic form. Since research productivity is measured
by the number of papers produced by the research project (controlled for
inputs), our dependent variable is necessarily discrete. Moreover, we only
have data on research projects that published at least one paper. For these
reasons, we use a zero-truncated NB regression model. An alternative would
be a Poisson regression, but that is not suitable in this case, since the equi-
dispersion hypothesis is strongly rejected at the 1% significance level. Since
the estimator of the NB regression coincides with that of quasi-maximum
likelihood, it is robust to misspecification of the distribution of the dependent
variable. That is, the NB regression yields a consistent estimator as long as
the specification of the conditional expectation of the regressand is correct.5
The unit of analysis for the H-3 model is the entire research project,
whereas the unit of analysis for the H-1 and H-2 models is the focal paper.
This is because the H-3 model is concerned with the effect of management
structure on the productivity of the entire project. In estimating the H-
1 and H-2 models, we restrict the observations to those cases where the
respondent is the first author of the focal paper, while for the H-3 model
we use observations for which the respondent was the researcher who took
5See Cameron and Trivedi (2005) for further discussion and Ding et al. (2010) for anapplication in a related context.
20
a central part in the research and contributed the most.6 Thus, the author
characteristics controlled in those two models are those of the first author,
while in the H-3 model they are those of the principal investigator. The
project funds variable is modified as the amount per paper for the H-1 and
H-2 models.
3.5. Estimation Issues
3.5.1. Sampling Bias
As a consequence of the survey method, one-third of the samples were
chosen from those researchers who wrote one of the top 1% highly cited
papers. Hence, our samples are not randomly drawn from the entire popu-
lation. We must consider this problem in order to obtain a consistent esti-
mator for the H-1 model, since the stratification depends on the regressand
(i.e., the number of citations).
A straightforward way to address this problem is to use the weighted least
squares method. In our regressions, we have two strata: highly cited papers
and others. Each observation i is weighted by the ratio between the popula-
tion frequency and the sample frequency of the stratum to which i belongs.
The weights are 0.032 for the highly cited papers and 1.433 for the others.
Under reasonable regularity conditions, this weighted least squares estimator
is consistent and asymptotically normally distributed (Wooldridge, 2010a,
2010b). Moreover, a consistent estimator of covariance matrix is obtained
by slightly modifying White’s (1980) heteroskedasticity-consistent covari-
ance matrix.
6In the H-1 and H-2 models, we excluded samples who are the corresponding but notfirst author of the focal paper to control the main author’s characteristics. About 30% ofthe sample was dropped after this selection.
21
3.5.2. Selection Bias
A number of anecdotes have casually reported that serendipity was im-
portant for successful research. Clearly, this cannot be taken as evidence
for a general tendency, since dramatic and successful anecdotes tend to be
highlighted and selected. As noted above, this paper adopts a definition of
serendipity, following Stephan (2010), as neutral as possible to the conse-
quential value of the serendipitous finding. Our survey contains many cases
in which scientists report that their finding was serendipitous in the sense
that it answered questions not yet posed, and yet the value of the finding,
measured by the number of citations, was not that large. By comparing
serendipitous findings to intentional findings with both successful and un-
successful findings, this paper aims to estimate the difference in the potential
values of findings that are found intentionally and accidentally from the per-
spectives of scientists. The neutral definition of serendipity enables us to
avoid the issue of selection bias that arises from overlooking the cases of
serendipitous but not highly valued findings.
3.5.3. Endogeneity Bias
The H-1 model concerns the population difference in the value of findings
between serendipitous findings and intentional ones. A regression of value on
serendipity may not yield a consistent estimator for this difference, however,
if the value difference affects the observation of serendipity in our data. In
other words, the estimate is biased if serendipity is an endogenous variable.
The serendipity observed in the data can be affected by the consequen-
tial value of the finding in the following sense. The value of the finding and
the serendipitous event are observable only conditional on the publication
of the finding. Moreover, a research team pursues publication only when
22
it considers it to be valuable. Hence, the observed citation rate might re-
flect not only the intrinsic value of a finding uncovered by a serendipitous
event, but also the research team’s evaluation of that discovery. There can
be two directions of endogenous effects from the quality of the finding to its
serendipity. The first possibility is that the researcher is less familiar with
the topic pertaining to the serendipitous finding than the topic originally
pursued; hence, the finding seems more novel to the researcher, who overes-
timates the value of the serendipitous finding. The second possibility is that
the researcher pursues their serendipitous finding only if it is highly valu-
able. This is because diverting the direction of research from the original
plan seems undesirable or risky. While we cannot determine which effect of
endogeneity is dominant ex ante, both effects imply that a simple regression
would result in underestimation or overestimation of the effect of serendipity
on research quality. Indeed, in the estimation of the H-1 model, a variable
addition test rejects the hypothesis that serendipity is an exogenous variable
at the 1% significance level (see Table 4).
We use instrumental variables to deal with the endogeneity bias. The
instrumental variables must correlate with the existence of serendipitous
findings, and they must not affect the ex ante evaluation of the findings. To
satisfy this criterion, we use two instrumental variables, skill diversity and
inter-lab community.
It is plausible that serendipity correlates with these two variables, since
complementarity in knowledge and skills are key to enhancing creativity.
For example, Heinze et al. (2009) pointed out that communication with spe-
cialists who have different knowledge or skills constitutes one of the most
important factors in inspiring a researcher’s creativity. Furthermore, we
assume that our instruments do not directly affect the ex-ante evaluation
23
of the serendipitous finding for the following three reasons. First, by con-
struction, these two instruments characterise the entire project rather than
the focal research paper. Second, even if the instruments affect the qual-
ity of the findings, the effect occurs mainly through enhancing creativity.
For example, skill diversity, which indicates whether communication with
researchers with different skills was important for conceiving the research
project, improves the value of findings by increasing the chance of valuable
serendipitous findings. In fact, we find that the variation of number of cita-
tions explained by skill diversity through serendipity in our 2SLS estimate
almost exhausts the correlation between number of citations and skill diver-
sity. Third, we conducted Sargan and Bassman over-identification tests to
see whether these two instruments are exogenous.7 The tests did not reject
the hypothesis that all the instruments are exogenous at the 20% significance
level. See Table 4 for the detailed results.
4. Results
4.1. Baseline Estimates
Table 4 summarises the estimation result for the H-1 model. We observe
that serendipity has a positive correlation with the logarithm of citation
counts at the 5% significance level in the 2SLS regression. This confirms our
hypothesis H-1: the findings through serendipitous events exhibit a higher
value in terms of the number of citations than findings which were made
intentionally. Note that this effect is insignificant in the OLS regression.
Namely, our sample does not exhibit a significantly positive correlation be-
tween serendipitous events and greater citation counts directly. In other
7Since our sample is drawn from two groups, highly cited and others, with differentsampling rates, we conducted over-identification tests for each subsample.
24
words, there are many observations which report serendipitous events and
low citations. When the serendipitous events are instrumented by exoge-
nous variables, however, a positive effect of serendipity is identified. We
interpret this estimation result as follows. Researchers tend to overestimate
the value of a serendipitous finding, and thus pursue the publication of the
finding with a less stringent criterion than intentional findings. As a result,
relatively more publication of serendipitous findings with low value are ob-
served, which masks the intrinsic positive quality of serendipitous findings.
However, the estimate may be open to other interpretations.
Among the control variables, we note that the existence of a threat from
competitors and the number of past publications of the first author have sig-
nificant positive effects. This is natural, as the competitor threat proxies the
potential value of the research topic, and the publication record represents
the researcher’s ability. We observe that the project size is negatively re-
lated to the quality on average. We robustly find that university researchers
exhibit less citation counts, which may reflect the fact that more of them
publish in a field with fewer researchers, than do researchers in industry.
The country effect shows that the average number of citations is higher in
the U.S. than in Japan.
The H-2 model examines the connection between management structure
and serendipity. In Table 5, the left column exhibits the result of the base-
line probit regression for H-2. We observe a positive correlation between
management–research integration and serendipity. The baseline model is a
heteroskedastic probit, because the homoskedastic specification is statisti-
cally rejected at the 1% significance level. The variance of the probit index
depends on project duration, which is likely to be an exogenous variable in
this model.
25
Table 5 also indicates that serendipity is reported more often when the
researchers were more engaged with an inter-laboratory community (a re-
search community beyond their own labs) or when the conception of the
project benefited from communication with researchers with different skills.
This result justifies our use of these variables as instrumental variables in the
estimation of the H-1 model. In addition, we note from the country dummy
that the Japanese respondents reported serendipity more often than the U.S.
respondents.
The regression results of the H-3 model are shown in Table 6. Our
hypothesis is that the separation of management from research increases
research productivity, measured in the number of publications generated by
the entire project. The first column shows that the separation exhibits a
positive effect on the number of publications. When the interaction with
project size is included as in the second column, separation has a negative
coefficient, while the interaction term exhibits significant positive effects.
Thus, the estimates imply that the separating management style yields a
higher productivity for a large project team. This result conforms to our
hypothesis. Quantitatively, the baseline estimate predicts that the adoption
of a separating management style increases the number of publications for a
project with more than three researchers. The average increase is 1.9 for a
project with 6 researchers and 4.8 for a project with 10 researchers, whereas
the projects with 6 and 10 researchers correspond to the median and the
75th percentile in project size, respectively. Considering that the median
number of papers published by a project is 7, the increase of publications
by adopting a separating management style seems non-negligible.
The coefficients for the control variables indicate that the number of pa-
pers generated by the project is large when the project fundsing is large,
26
when the project duration is long (though the effect is attenuated by the
squared term), when threatening competitors exist, and when the principal
investigator has a good publication record or a Ph.D. Projects led by univer-
sity researchers generate fewer papers than the other projects, indicating the
same pattern as for the H-1 model, where university researchers receive less
citations. U.S. projects also tend to generate fewer papers than Japanese, in
contrast with the result in H-1, which showed that U.S. researchers received
more citations than Japanese.
4.2. Robustness Checks
We conducted various robustness checks. First, we tested the validity of
our choice of instrumental variables. Three different choices were examined:
skill diversity (labeled “H-1-1”), knowledge diversity (labeled “H-1-2”), and
knowledge diversity and inter-lab community (labeled “H-1-3”). Table 7
shows the results. All three models suggest that serendipity and the number
of citations are positively correlated, so that our hypothesis is maintained.
Secondly, for the H-2 model, we tested different specifications of the error
term. The estimation for the homoskedastic probit is shown in the right
column of Table 5. We observe that the estimated coefficients are qualita-
tively similar to the baseline results. We also confirmed that an alternative
specification with variance depending on country as well as project duration
generates a similar pattern. Finally, we restricted the sample in the H-3
model to the projects which clearly stated their management structure, by
dropping the observations stating that management was not necessary or
choosing “other” in the survey. The estimates in Table 6 show that all the
major results are still qualitatively unchanged. Moreover, dropping the out-
liers (top 1%) of project size, published papers, and past publications in the
27
H-3 model did not alter our estimation result qualitatively.
5. Conclusion
This paper investigated the effects of division of labour between manage-
ment and research on serendipity and productivity in scientific research. The
major estimated results reveal the following three points. First, the estima-
tion shows that on average serendipity brings about better research quality.
Much of the anecdotal evidence has suggested serendipity plays a critical
role in science. The estimation from the survey empirically demonstrates
that serendipity is a key feature of science not only in great discoveries but
also in science in general. This finding suggests the importance of a man-
agement that gives scientists the flexibility to pursue a serendipitous finding
when they face the unintended and unexpected. Second, the integration of a
managerial role with a leading research role has a positive effect on serendip-
ity. Following the discussion of the division of labour and its coordination
costs, it can be interpreted that integration reduces the coordination costs
between management and research and provides scientists with flexibility in
research. When a scientist finds unintended and unexpected findings, the
findings are usually still too crude and uncertain to be articulated. Thus, if
the front-line scientist is delegated decision-making in the research project,
they can fully desterilise any uncodified tacit knowledge and use managerial
resources in the context of the actual situation. However, if a hierarchical
managerial role, which increases the information asymmetry and incommen-
surability between management and actual research, is implemented in a top
down fashion, it rebuffs research efforts to pursue a serendipitous finding.
Thus, the serendipitous findings based on ground level intuition are seldom
pursued. Thirdly, the results show that if a project is managed not by a
28
scientist who actually leads the scientific research, but by a person who spe-
cialises in research management, it increases the productivity of the research
when measured by the number of papers. These three findings imply that
there is a trade-off between pursuing serendipity and achieving research ef-
ficiency in science, via who plays the managerial role and who the leading
research role.
Returning to the example of penicillin, the findings of the paper suggest
that Fleming would have faced difficulties in changing his original research
plan to pursue his serendipitous finding if he had been working in a large
laboratory and his research had been led by a competent project manager.
In other words, Fleming would not have pursued the serendipitous finding,
but he would have delivered more papers concerning the original research
project if a managerial role had been played by a specialised director.
Serendipity plays an essential role in discoveries not only in science, but
also in technology, management, business practices, art, and daily life (Ja-
cobs, 2010; Svensson and Wood, 2005; Van Andel, 1992). The findings
of this paper have implications for corporate R&D and university research
administrators in particular. First, these results about the effects of the
division of labour between research and management on serendipity and
productivity in science are consistent with the contingency theory of firms
between the complexity of environment (e.g., demand, strategic positioning,
and technology) and their organisational structure (Burns and Stalker, 1961;
Lawrence and Lorsch, 1967; Scott, 1981), which have indicated that decen-
tralised and less formalised management allows a high degree of flexibility.
This is suitable when an organisation faces many exceptional problems and
problem solving is not easy (Perrow, 1967; Woodward, 1965). It suggests
that if corporate R&D is involved with embedded and uncodified knowledge
29
and if the firm needs alert responses to rapidly changing demand and supply
conditions, greater autonomy should be given to the R&D unit (Birkinshaw
et al., 2002). This may explain why it is difficult for corporate R&D over-
seen by a central business director to profit from serendipitous findings in
the laboratory. The findings of this paper suggest that decisions should be
made where the important information is gathered and knowledge is created
if unexpected findings are important.
Secondly, the findings have implications for university research adminis-
trators. Since the size of research projects is increasing and the competition
for priority of discovery in science is becoming fiercer, university research
administrators who are responsible for planning and managing research ac-
tivities and promoting research outcomes are serving very important roles in
science (Kaplan, 1959). According to the findings of this paper, the research
outcomes of the team depend highly on the extent to which the leading re-
search role and the managerial role are divided in the team. Operational
administrators are usually trained to complete the project’s goal. In fact,
they attempt to manage in a way that will eliminate uncertainty in their af-
fairs so that they can meet budgets and target deadlines (Udwadia, 1990). If
a managerial decision is made by such an operational research administrator,
it is highly likely that they would adhere closely to the initial research plan
even when scientists in the team encounter something serendipitous. Our
results do not suggest that the division of labour in research and manage-
ment is always inappropriate in science. The findings of this paper highlight
the way that the separation of research from managerial role allows the team
to achieve higher productivity as measured by the number of publications
for a given level of inputs. It would also work well if the research project
aims to make a thorough investigation of the possible combinations of ma-
30
terials with a door-to-door-check method, for example. Additionally, it is
quite important for a university research administrator to fully understand
the nature of discovery in science and the trade-off between serendipity and
productivity in science via who plays the managerial role and who the lead-
ing research role in research management (Kaplan, 1959; Kulakowski and
Chronister, 2006).
For the purpose of thinking about future research, it is important to note
three limitations to the present study. The first is related to the time scope
of the research project. A research project does not stand alone for a scien-
tist. Instead there is usually a sequence of continuous and related projects.
Even when the project is defined by the respondent themselves, as an entire
body of research that generated the focal paper and related papers, there
still remains, to some degree, a continuity of the research projects. This
problem is also related to the measurement of the quality of the research.
The scientists’ survey, which aimed to explore the nature of high performing
research projects compared with that of randomly selected research projects,
measured the quality of a research project by the number of citations that
the focal paper of the project received. Even though the measurement of
research quality by the number of citations has been widely used, there are
debates over the validity of this method. One of the issues related to the
present paper lies in the base year of the citation. Although it is necessary
to set a certain base point for measuring the number of citations, one might
always question whether some research might become highly valued after
the data cut-off point, for instance, due to the advancement of complemen-
tary technology. Since the scientists’ survey adopted a relatively short data
cut-off point because it aimed to distribute the questionnaires properly to
the corresponding authors, there is a possibility that this data cut-off point
31
underestimates the quality of research that takes a long time to be valued. In
addition, we used cross-sectional data, rather than a panel dataset. There-
fore, it should be noted that our empirical results are patterns of associations
between variables. Even though we carefully introduced instrumental vari-
ables in order to eliminate endogenous bias and tested our hypotheses based
on the patterns of correlations in the regression models, it is important to
collect panel data for deducing causal relations. These time scope issues are
an endemic problem in the study of science.
The second point is related to the country effects. The main aim of this
paper was to explore the effects of the division of labour between manage-
ment and research in science on serendipity and publication productivity.
Thus, although this paper introduces a country dummy variable to control
for country effects, it did not address the international comparison of the
management of science. However, as the estimations show, the country vari-
ables show significant effects on serendipity and publication productivity. It
is reasonable to assume that the ways in which scientific research is organ-
ised and managed are different across countries. Since the sample size of the
study did not allow robust estimation in the international comparison, it is
potentially important to collect a larger sample of the management of science
across countries so that one can make detailed international comparisons.
The third point is linked to the quality of the project managers. A
key result suggests that if scientific research is bureaucratically controlled
in a research organisation, serendipitous encounters will not be realised. In
other words, even when a managerial role and a leading research role are
played by different people, serendipity will be realised if a manager shares
tacit and domain-specific knowledge with leading scientists and understands
the nature of scientific discovery. This paper presupposes a certain degree of
32
incommensurability, which was proposed by “Kuhnian paradigm arguments”
(Kuhn, 1970) between a manager and leading scientists. However, the degree
of incommensurability depends on a manager’s expertise and capabilities.
Since the scientists’ survey does not allow the investigation of a manager’s
capabilities, this paper does not explore the quality of managers in a research
organisation. Organisations for university research administrators such as
SRA (Society of Research Administrators) and NCURA (National Council of
University Research Administrators) in the U.S., and ARAM (Association
of Research Managers and Administrators) in the U.K., were established
in the 1960s. Not only these organisations but also governments (e.g., the
Development of a Research Administration System program launched by the
Ministry of Education, Culture, Sports, Science, and Technology in Japan)
are beginning to understand that a managerial role should be played by a
specialist who can share tacit and domain-specific knowledge with leading
scientists: scientists could then focus on large-scale research projects, which
could have the managerial flexibility for realising serendipitous encounters.
The previous literature on how scientists with different sets of expertise
and paradigms communicate has indicated that scientists communicate in
groups called “trading zones,” where they can agree on the rules of exchange,
share the same learned languages, and share tacit knowledge (Collins et al.,
2007; Galison, 1997, 1999). However, since the extent to which managers
and scientists can reduce the degree of incommensurability depends on a
manager’s ability, it is important to explore the manager’s expertise and
capabilities for the research outcome in detail.
33
Table 1: Definitions of Variables
Variable Definitionnumber of citations Cumulative number of citations in 2009.serendipity Equals 1 if their research output found the answers to ques-
tions not originally posed.published papers The total number of refereed publications produced by the
research project.integration Equals 1 if the researcher executed the central part of the
research and contributed the most to the research output,and at the same time, took a leading role in the researchmanagement, designing the research project, organising theresearch team, and/or acquiring research funds.
separation Equals 1 if the researcher executed the central part of theresearch and contributed the most to the research output,but on the other hand, played no managerial role.
project size Sum of the number of collaborative researchers (includingcoauthors), graduate students, undergraduates, and techni-cians involved in the project.
project duration Years between the launch of the research project and thelatest publication by the project.
project funds The total sum of research funds prepared for the project.skill diversity Equals 1 if the researcher states that communication with
researchers who have different research skills was importantfor conceiving the research project.
inter-lab community Equals 1 if the researcher built a research community be-yond own laboratory.
knowledge diversity Equals 1 if the researcher states that communication withvisiting researchers or postdoctoral researchers was impor-tant for conceiving the research project.
competitor threat Equals 1 if the researcher considered the possibility of com-petitors who may have obtained priority in the researchresults.
past publications The number of refereed papers that the researcher publishedin the three previous years.
years in paper Years between the launch of the project and the publicationof the focal paper.
age Respondent’s age at the time of survey.PhD Equals 1 if the researcher had a Ph.D. or equivalent degree.award Equals 1 if the researcher received a distinguished paper
award or a conference award.university Equals 1 if the researcher works for universities.country Equals 1 for the respondents in the U.S. and 0 for respon-
dents in Japan.theory Equals 1 if the researcher specialised in theoretical work.experiment Equals if the researcher specialised in experiments.
34
Table 2: Summary Statistics
H-1 Model (Obs.= 1629) H-3 Model (Obs.= 1892)Variable Mean SD Min Med Max Mean SD Min Med Max
Standard errors in parentheses∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01Note: Over-identification tests are conducted separatelyfor N (normal papers) and H (highly cited papers).
37
Table 5: H-2 Model (Effect of Management Structure on Serendipity)
Question: Please indicate which of the following best describes your role in the man-agement of the research project.
Response RatesAnswers Highly Cited Normal(1) A leading role in the research management, design-ing the research project, organising the research team,and/or acquiring research funds
70.9% 69.2%
(2) A member of the research management, but a roleless than that of the leader
14.1% 14.8%
(3) No managerial role 7.2% 5.8%(4) Management was not necessary 5.8% 8.0%(5) Other 2.1% 2.3%
Question: Please indicate which of the following best describes your role in the re-search implementation.
Response RatesAnswers Highly Cited Normal(1) I executed the central part of the research and con-tributed the most to the research output
64.4% 65.5%
(2) I took part in the central part of the research, but mycontribution was not as substantial as that of the centralresearcher
20.8% 21.9%
(3) I implemented the research under the guidance of theabove members
2.1% 3.0%
(4) I contributed to the research through the provisionof materials, data, equipment, or facilities
2.7% 2.8%
(5) Other 10.0% 6.8%
41
References
Adams, J. D., Grant, C. B., Clemmons, R. J., Stephan, P. E., 2005. Scientific
teams and institutional collaborations: Evidence from U.S. universities.
1981–1999. Research Policy 34 (3), 259–285.
Agrawal, A., Goldfarb, A., 2008. Restructuring research: Communication
costs and the democratization of university innovation. American Eco-
nomic Review 98 (4), 1578–1590.
Allen, T. J., 1977. Managing the Flow of Technology: Technology Trans-
fer and the Dissemination of Technological Information within the R&D
Organization. MIT Press, Cambridge, MA.
Andrews, F. M., 1979. Scientific Productivity: The Effectiveness of Research
Groups in Six Countries. Cambridge University Press.
Arora, A., Gambardella, A., 1994. The changing technology of technological
change: General and abstract knowledge and the division of innovative
labour. Research Policy 23 (5), 523–532.
Barber, B., Fox, R. C., 1958. The case of the floppy-eared rabbits: An
instance of serendipity gained and serendipity lost. American Journal of
Sociology 64 (2), 128–136.
Barnard, C. I., 1938. The Functions of the Executive. Harvard University
Press, Cambridge, MA.
Becker, G. S., Murphy, K. M., 1992. The divison of labor, coordination costs,
and knowledge. Quarterly Journal of Economics 107 (4), 1137–1160.
42
Birkinshaw, J., Nobel, R., Ridderstrale, J., 2002. Knowledge as a contin-
gency variable: Do the characteristics of knowledge predict organization
structure? Organization Science 13 (3), 274–289.
Brown, J. S., Duguid, P., 1991. Organizational learning and communities-
of-practice: Toward a unified view of working, learning, and innovation.
Organization Science 2 (1), 40–57.
Burns, T. B., Stalker, G. M., 1961. The Management of Innovation. Tavis-
tock Publications, London.
Cameron, C. A., Trivedi, P. K., 2005. Microeconometrics: Methods and
Applications. Cambridge University Press.
Chesbrough, H., 2003. Open Innovation: The New Imperative for Creat-
ing and Profiting from Technology. Harvard Business School Press, Cam-
bridge, MA.
Collins, H., Evans, R., Gorman, M., 2007. Trading zones and interactional
expertise. Studies in History and Philosophy of Science 38, 657–666.
Collins, H., Harrison, R. G., 1975. Building a tea laser: The caprices of
communication. Social Studies of Science 5 (4), 441–450.
Dasgupta, P., David, P., 1994. Toward a new economics of science. Research
Policy 23 (5), 487–521.
Ding, W. W., Levin, S. G., Stephan, P. E., Winkler, A. E., 2010. The
impact of information technology on academic scientists’ productivity and