Journal of the American Society for Information Science and Technology - 2007 - Birnholtz - When do researchers collaborate

This exploratory study compares two approaches to
understanding the “collaboration propensity” of individ-
ual researchers. On the one hand, social comparisons of
disciplines would suggest that collaboration is a function
of orientation toward individual versus collective respon-
sibility for discovery. A contrasting approach would hold
that collaboration depends on the work researchers are
engaged in—when it is useful to collaborate, they will do
so regardless of the social climate. Results presented
here suggest that this latter approach is potentially more
powerful but that there are complexities in measurement
and operationalization that urge a more nuanced treat-
ment of collaboration propensity.
Introduction
Collaboration in research has been a topic of significant
recent interest as the scale of research projects has increased
(Galison & Hevly, 1992; Price, 1963), the social networks of
researchers have broadened in scope (Wagner & Leydesdorff,
2005), and there has been substantial investment in improving
networking technologies that facilitate high–end computing
and work in geographically distributed groups (Finholt &
Olson, 1997; Hara, Solomon, Kim, & Sonnenwald, 2003;
Hesse, Sproull, Keisler, & Walsh, 1993; Newell & Sproull,
1982). Moreover, recent research and reports suggest that we
are presently at a crucial juncture in the development and adop-
tion of “cyberinfrastructure” technologies for “eScience”
activities (Atkins et al., 2003; Nentwich, 2003). Even as access
to these technologies spreads, effective collaboration remains
difficult and our understanding of it is incomplete. More
specifically, collaboration must occur within a work and
reward structure that is largely focused on individual achieve-
ment and reputation (Kennedy, 2003; Whitley, 2000), and
coordination difficulties can impact productivity and effec-
tiveness (Cummings & Kiesler, 2005).  
While there is a desire to address these difficulties
through the improvement of cyberinfrastructure and the
research environment, such action requires an improved
understanding of collaboration itself, and this has been
reflected in recent calls for social scientists to explore these
issues (e.g., Mervis, 2005). More specifically, it is important
to understand how research collaboration works and what
makes it desirable to individual researchers. With an under-
standing of what makes collaboration desirable and useful, it
becomes possible for funding agencies to invest limited
resources in collaborations that are most likely to succeed
and to foster conditions under which other successful collab-
orations are likely to take shape.
Decisions about collaboration at the individual level have
been shown to depend on a range of factors, including the
prior experience of participants (Hara et al., 2003), institu-
tional constraints (Landry & Amara, 1998), the availability
of “attractive” collaborators in terms of influence or unique
skills (Bozeman & Corley, 2004), entrepreneurial aspira-
tions (Oliver, 2004), and the need for access to special data or
research equipment (Beaver, 2001; Kouzes, Myers, & Wulf,
1996; Melin, 2000). Some of these studies have also pointed
out differences between research fields in the amount and
types of collaboration observed (Bozeman & Corley; Melin),
but they have not made clear the extent to which these differ-
ences result from variation in the nature of work being con-
ducted versus social differences among fields. The present
study is an exploratory move in this direction. 
In the article that follows, two perspectives on collabora-
tion are considered and compared. On the one hand, social
and cultural comparisons of researchers in different disci-
plines (e.g., Collins, 1998; Hargens, 1975; Knorr Cetina,
1999) suggest that observed differences in collaboration
behavior may be a function of socialization and particularly
the extent to which individual versus collective entities are
emphasized in the attribution of credit for discoveries (Knorr
Cetina, 1999). On the other hand, it is possible that these
cultural distinctions are epiphenomenal and that observed
differences in collaboration behavior result from differences
in the nature of work that scientists in different fields are
JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE AND TECHNOLOGY, 58(14):2226–2239, 2007
When Do Researchers Collaborate? Toward 
a Model of Collaboration Propensity
Jeremy P. Birnholtz 
Department of Communication, Cornell University, 310 Kennedy Hall, Ithaca, NY 14853. 
E-mail: jpb277@cornell.edu
Received December 7, 2006; revised February 15, 2007; accepted February
16, 2007
© 2007 Wiley Periodicals, Inc. •Published online 26 September 2007 in
Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/asi.20684

JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE AND TECHNOLOGY—December 2007 2227
DOI: 10.1002/asi
engaged in.In the quantitative data presented here, this latter
approach turns out to more powerful in understanding col-
laboration, though the qualitative data do suggest that social
factors have some more subtle effects. 
Research Collaboration: Background
Much research on collaboration to date has used biblio-
metric data (e.g., Cronin, Shaw, & La Barre, 2003; Price,
1963), sometimes using social network analysis as well
(e.g., Newman, 2001; Wagner & Leydesdorff, 2005). While
such techniques are useful for capturing “macro level” phe-
nomena, Katz and Martin (1997) point out that bibliometric
data are not always valid proxies for actual collaborative
behavior. A few prior studies have examined collaboration
at the “micro level.” Hagstrom (1965) and Melin (2000), for
example, suggest that collaboration is driven by the need 
for access to the instrumentation and expertise required to
answer questions of interest. Others have investigated the
relationship between spatial propinquity and collaboration,
suggesting that collaboration is more likely to occur
between closely collocated researchers than those who are
geographically disparate (Allen, 1977; Hagstrom, 1965;
Kraut, Egido, & Galegher, 1990). At the same time, how-
ever, Wagner and Leydesdorff (2005) show the growth of
international collaboration networks, so propinquity is
clearly not the only factor. Hara, et al. (2003) developed a
preliminary framework for assessing the nature of collabo-
rations and the conditions of their formation; but, none of
these studies provide a sense of the relative weights and
range of factors that influence what will be referred to here
as the “collaboration propensity” of individuals.
“Collaboration propensity” is defined for the purposes of
this study as the likelihood of an individual researcher
engaging in collaboration at a particular point in time and
with regard to current research interests1. Existing literature
suggests that collaboration propensity is comprised primar-
ily of two classes of attributes. First is whether researchers
believe that collaboration will provide them with access to
expertise, apparatus, data sets, or other resources necessary
in answering research questions of interest (Beaver, 2001;
Hagstrom, 1965). Second is the extent to which researchers
perceive collaboration as a component of building an individ-
ual reputation and establishing a viable career path (Whitley,
2000).  Collaboration propensity will serve as the dependent
variablein the present study, and it will be measured using
five scale items developed for this study. (See Appendix A.)
The remainder of this article explores the measurement and
prediction of collaboration propensity, both in terms of
quantitative survey results and qualitative interviews that
highlight the complexity of issues involved in thinking care-
fully about collaboration.
Predicting Collaboration Propensity: 
Hypotheses
In this exploratory study, the predictive power of several
independent factors, each based on one of the approaches
being compared, is assessed with regard to collaboration
propensity. 
Individual versus Collectivist Orientation
Knorr–Cetina (1999) and Collins (1998) have both
pointed to social differences centered around individual ver-
sus collective responsibility for discovery in research.
Knorr–Cetina, in particular, highlights differences between
high–energy physics (HEP) and molecular biology, arguing
that reputation in the latter field is strongly attributed to indi-
vidual researchers and laboratory leaders who compete
fiercely to be first author on papers and accrue reputational
“credit,” while the individuals she studied in HEP tended to
be subsumed by large collaborative entities. This leaves
open the question, however, of whether these social differ-
ences actually correlate with different attitudes toward col-
laboration. This question will be explored here by focusing
on two issues. 
Scientific Competition
Science is generally viewed as competitive in that individ-
ual researchers compete intensely in an “economy of reputa-
tion” to be the first to make unique and groundbreaking
discoveries (Whitley, 2000). As Knorr–Cetina (1999) has
pointed out, however, there is substantial variation in how this
competition plays out. Concerns about competition and fears
about being “scooped” have however been shown to impact
researchers’ willingness to share data (Zimmerman, 2003),
adopt database systems for sharing resources even with their
known collaborators (Birnholtz & Bietz, 2003), and discuss
research in progress with colleagues (Blumenthal, Campbell,
Anderson, Causino, & Louis, 1997; Hagstrom, 1974; Walsh &
Hong, 2003). In this study, scientific competition will be mea-
sured with four questionnaire items (see Appendix A) based
on those used by Walsh and Hong (2003) that ask about the
extent to which scientists are concerned about discussing their
results with colleagues, their concerns about being scooped,
and the perceived importance of winning individual prizes
and widespread recognition. It is expected that,
H1: There will be a negative relationship between collabo-
ration propensity and the perceived level of scientific
competition.
Ease of Collective Credit Attribution
Another way to consider individual versus collective
responsibility is the ease with which credit can be assigned
to groups of researchers. Significant contributions to
research projects are typically acknowledged via authorship
1This is not being treated here as a persistent character trait, though it is
certainly possible that there are persistent individual attributes that influence
collaboration propensity.
 15322890, 2007, 14, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/asi.20684 by Wuhan University, Wiley Online Library on [07/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License


on scholarly publications, but there are differences in how
this credit is assigned. Some fields of study, for example,
place significant value on being the first author listed in
an article (Engers, Gans, Grant, & King, 1999; Laband &
Tollison, 2000); but, when there are many contributors, it
may be unclear where different authors will be listed in an
article. In other fields, such as HEP, all contributors are listed
alphabetically on all articles. (See Birnholtz, 2006 for a
detailed discussion of authorship practices in HEP.) Ease of
credit attribution will be defined here as the ease with which
one’s coauthors can be determined when engaged in collab-
orative research. It will be measured using two scale items
developed for this study. (See Appendix A.)
Where researchers can easily determine at the start of a
collaborative project how they will receive credit for their
contributions, we might expect to see differences in collabo-
ration propensity. Thus,
H2: There will be a positive relationship between the ease of
collective credit attribution and collaboration propensity.
Considering Attributes of Scientists’ Work
Fuchs (1992) proposes a theory of scientific production
that focuses not on the social aspects of science but on the
nature of the work in which scientists are engaged from an
organizational and resource coordination standpoint. An
approach to collaboration propensity rooted in Fuchs’s model
would hold that social differences among research disciplines
are epiphenomenal, and that it is the attributes of work itself
that are important in understanding collaboration. Such an
approach would suggest that where collaboration is useful or
necessary in answering interesting research questions, we
should expect to see an increase in collaboration propensity.
The remainder of this section enumerates a set of constructs
fundamentally rooted in Fuchs’s model and outlines a second
set of hypotheses.
Field–Wide Focus
Fuchs (1992), Whitley (2000), and Hargens (1975) all
point to the heterogeneity of methodological approaches in a
field as an important determinant of how progress in that
field takes place. Field–wide focus is here defined as a mea-
sure of homogeneity with regard to research questions and
methods within a field of research. It will be measured using
three scale items developed for this study that ask respon-
dents about the perceived level of agreement on methods
and important research questions. (See Appendix A.) This
notion is based on Fuchs’ concept of task uncertainty and
Hargens’ concept of normative integration. Focus is low
where work in a field uses a variety of methodological
approaches to many loosely related research problems. This
stands in contrast to fields such as HEP, where the important
questions and appropriate research methods are widely
agreed upon. Focus is likely to impact collaboration propen-
sity in that people in highly focused fields will be able to
work together more effectively and be more likely to find
like-minded collaboration partners. 
H3: There will be a positive relationship between focus and
collaboration propensity.
Resource Concentration
Fuchs (1992) and Whitley (2000) both discuss the notion
of mutual dependence as the extent to which researchers in a
field depend on each other for access to sufficient quantities of
scarce resources (e.g., funding, use of an apparatus) to be able
to answer interesting research questions. Areas in which
experimental and financial resources are concentrated at a
small number of locations or controlled by a small group of
researchers will be referred to here as having a high degree of
“resource concentration.” This will be measured using two
scale items developed for this study that ask respondents about
the extent to which research in their area requires expensive
equipment or large amounts of funding. (See Appendix A.)
Where resource concentration is high, researchers are likely to
be dependent on others for access to these scarce experimental
and financial resources (Thompson, 1967). In turn, collabora-
tion propensity will likely increase.
H4: There will be a positive relationship between the per-
ceived level of resource concentration and collaboration
propensity.
Agreement on Quality
Another dimension of Fuchs’ (1992) concept of hetero-
geneity is the extent to which there is agreement on what
constitutes high quality research. Hargens (1975) studied
several indicators of such agreement, including the rejection
rates of journals, lengths of student theses, and the extent to
which there were perceived hierarchies of journals and insti-
tutions. Agreement on quality is defined here as a measure of
the extent to which individual researchers perceive wide-
spread agreement on what constitutes quality research in
their field of study. It will be measured here using five scale
items developed for this study that ask respondents about
how they assess their peers’ work and about the perceptions
of hierarchies of journals and institutions. Just as a high
degree of focus seems likely to affect the probability of find-
ing like-minded collaborators, widespread agreement on
quality also seems likely to affect researchers’ ability to find
collaborators with whom they can work successfully. Thus,
it is expected that
H5: There will be a positive relationship between agreement
on quality and collaboration propensity.
Need for and Availability of Help
While Fuchs’ theory does not directly address communica-
tion and interaction among scientists, his discussion of coor-
dination problems does suggest the importance of effective
2228 JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE AND TECHNOLOGY—December 2007
DOI: 10.1002/asi
 15322890, 2007, 14, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/asi.20684 by Wuhan University, Wiley Online Library on [07/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License


JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE AND TECHNOLOGY—December 2007 2229
DOI: 10.1002/asi
interaction and coordination between researchers. As suggested
by research on coordination (e.g., Malone & Crowston, 1994;
Van De Ven, Delbecq, & Koenig Jr., 1976), these interactions
are most important when individual work procedures are not
routine and there is frequent uncertainty about how to proceed.
This study therefore focuses on the amount of help–seeking
behavior in which researchers engage on a day–to–day basis, as
an indicator of frequent interactions and shared methodological
interests. This is measured here using four scale items, based on
those used by Van De Ven, et al., that ask respondents about the
routine nature of their work and the amount of assistance they
seek from colleagues. Where help is frequently needed and
readily available, we should expect there to be more opportuni-
ties for formal collaboration. 
H6: There will be a positive relationship between collabora-
tion propensity and the need for and availability of help.
Control Variables
In addition to the factors mentioned above, there are sev-
eral other demographic and basic factors that might influ-
ence collaboration propensity, such as field of study, field
tenure, the usage of Internet communication and collabora-
tion tools, and prior individual experience engaging in col-
laborative work. These factors will be included in this study
as control variables and are discussed in detail below.
Research Context and Methods
Research Context
Researchers in the academic disciplines of earthquake
engineering (EE), HEP, and neuroscience participated in
this study. These disciplines were carefully selected to
reflect diversity along the dimensions defined above. HEP,
as discussed above and by Knorr–Cetina (1999), has a
strong history of collaboration and collective credit attribu-
tion. Neuroscience, on the other hand, is more similar to the
molecular biologists in Knorr–Cetina’s study, as character-
ized by their strong emphasis on individual researchers
(1999). EE is in between these two fields in terms of individ-
ual versus collective orientation. These fields also represent a
diverse range of work attributes that are common to many
researchareas, which are described below.
High Energy Physics (HEP)
HEP has a rich history of collaborative experimental dis-
covery that is well chronicled elsewhere (Close, Marten, &
Sutton, 2002; Galison, 1997; Traweek, 1988). Experimental
investigations utilize high–energy accelerators that recreate
atmospheric conditions at the start of the universe.  By recre-
ating these conditions, physicists are able to generate
specific particles of interest that do not occur naturally under
more stable current atmospheric conditions. Large detectors
are used to track the behavior and existence of these parti-
cles by recording the energy “trails” left behind. Today’s
accelerators and detectors dwarf all other scientific instru-
ments. The Large Hadron Collider (LHC) at the European
Organization for Nuclear Research (CERN), the world’s
frontier laboratory, is an underground tunnel 27 kilometers
in circumference. The ATLAS (A Toroidal LHC Apparatus)
detector, one of two that will sit in the LHC when it is com-
plete in 2007, will be 20 meters in diameter and weigh 7000
tons. The human and organizational scale of this work is
similarly large, with approximately 2000 physicists from in-
stitutes all over the world involved in each of the two major
LHC experiments. While HEP is an admitted outlier, it is
nonetheless useful to study here for two reasons. First, the
community has been well–studied in the past, such that exist-
ing literature provides a useful starting point for this work
and facilitates comparison. Second, large, shared instru-
ments, facilities, and resources are increasingly common in a
range of fields (Galison & Hevly, 1992; Nentwich, 2003). 
Earthquake Engineering (EE)
Earthquake engineering is a field dedicated to the mitiga-
tion of earthquake risks by better understanding structural
response to seismic activity and using this knowledge to
modify architectural and design guidelines. Experimental
research typically takes place in large laboratories using
large–scale structural specimens, such as building or high-
way components, and specialized equipment, such as shaking
platforms and steel reaction frames (Sims, 1999). Simulated
seismic forces are exerted on specimens using hydraulic
actuators. Investigations generally involve teams of one or
two graduate students along with a faculty advisor, though
there have been some instances of larger projects involving
up to 10 faculty members. The specimen is generally instru-
mented with a large number of sensors from which numerical
data that capture specimen performance can be acquired and
analyzed. EE also exhibits traits common to many research
fields in that it is currently focused on single–investigator
research but where the scope of projects is slowly expanding;
and in that the apparatus required for research are scarce, but
there is increasing pressure from funding agencies to share
these resources with colleagues at other institutions.  
Neuroscience
Neuroscientists seek to improve treatment and prevention
of mental illness by understanding the detailed workings of
the brain. This is achieved primarily through laboratory
work, the statistical analysis of brain activity images and,
increasingly, via computational modeling of brain activity.
Laboratory work is generally done by individuals in tradi-
tional laboratory spaces and involves the analysis of brain tis-
sue from mice, primates, and humans, using gene microarrays
and other techniques. It is also common to use fluorescent pro-
tein tagging techniques and transgenic animals to isolate the
expression of particular genes related to specific traits and
illnesses. Where collaboration occurs in neuroscience, it is
 15322890, 2007, 14, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/asi.20684 by Wuhan University, Wiley Online Library on [07/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License


typically undertaken to share access to or bring multiple
forms of expertise to bear on the analysis and production of
different aspects of rare or expensive data sets. Moreover,
neuroscience is generally held to be an example of a 
wider set of biomedical fields of research that are highly
competitive and highly focused on individual researchers
(Kennedy, 2003; Knorr Cetina, 1999; “Who’d want to work
in a team?” 2003).
Methods
A combination of quantitative and qualitative methods
was used in carrying out this study. 
Quantitative 
A mail survey of 900 academic researchers affiliated with
universities and government laboratories in the United
States was conducted in April 2004. Three hundred
researchers were randomly chosen from professional direc-
tories in each of the three disciplines under investigation and
were mailed a paper questionnaire form (see Appendix A for
items) to complete and return by mail. The questionnaire
consisted of approximately 55 items, some of which were
developed specifically for this study and others borrowed
from prior work. All items had been examined and refined
during two preliminary pilot studies. 
Five hundred fourteen forms were returned, for a
response rate of 57%, but 133 were incomplete and had to
be discarded. There were no apparent differences between
the completed portions of the discarded forms and the com-
plete forms used in the analyses reported below. Response
was even across the three fields and 49% of the total
responses were from academic faculty. The remainder con-
sisted of graduate students (16%), postdoctoral researchers
(10%), academic research scientists (18%), and others (7%).
Respondents reported receiving their highest academic
degrees a mean of 14.1 years ago (SD /H1100512.3, n /H11005381), with
a range of 0 to 53 years. There were no apparent demographic
differences between respondents and nonrespondents.
For the purposes of the present article, only the respon-
dents holding a Ph. D. are considered in the analyses pre-
sented below. This is to simplify interpretation and eliminate
confounds related to measuring the collaboration behavior
of graduate students who, because of their position, may not
have the freedom to select collaborators or projects.
Prior to aggregation, the individual scale items compris-
ing each construct were checked for reliability via a confir-
matory factor analysis (Carmines & Zeller, 1979). In all
cases, the items were found to load significantly onto one
factor only. The individual items were then combined into
aggregate construct scores by summing the individual items.
Sums were used because there was no prior theoretical rea-
son to assume that any individual item would influence the
scale more than any other. The distributions of all items were
checked and found to be reasonably normal. Appendix B
shows the mean, standard deviation and range of values for
each of the constructs. Variables were also checked for
collinearity, which was not found to be a problem.
Qualitative Methods
As this is exploratory work, interviews were conducted to
aid in interpreting and explaining the results. Between May
2001 and November 2004, semistructured interviews lasting
20–60 minutes were conducted with 94 researchers in the
three disciplines being studied. Subjects were selected using a
combination of random and snowball sampling. Thirty–two
interviews were with physicists and conducted during a
10–week visit to CERN in Geneva, Switzerland.  Fifty were
with earthquake engineers during visits to 13 U.S. universi-
ties. The remaining 12 interviews were conducted with neu-
roscientists during visits to laboratories at one university and
over the telephone. 
Interviews were exploratory in nature at first. A similar
protocol was used in each discipline but was iteratively
refined as more was learned about the nature of work in each
discipline. Deliberate efforts were made to speak with indi-
viduals at various points in their careers, ranging from gradu-
ate students to senior faculty and from a range of institutions. 
Analysis of qualitative data consisted of careful reading
and rereading of interview transcripts both before and after
analyzing the quantitative data, while keeping in mind the
variables described above. As themes began to emerge, tran-
scripts were iteratively coded to reflect these themes using
established qualitative methods (Miles & Huberman, 1994).
These themes were then used to draw out the illustrations
presented below.
Results
In this section, quantitative and qualitative results will be
presented. Statistical analysis techniques are used to test the
hypotheses presented above, and qualitative data are used to
explain these results in greater detail.
Statistical Analyses
To analyze the survey data, a series of nested OLS
regression models were used in which the independent factors
were regressed on collaboration propensity (Neter, Kutner,
Nachtsheim, & Wasserman, 1996). The first model used basic
demographic and control variables, while subsequent vari-
ables were added one at a time (See Table 1.) The best fitting
model with the largest number of explanatory factors was
Model 4, with an adjusted R2 value of .53 (p /H11021.01). Model 5
is shown only to illustrate that adding additional factors added
no power.  Moreover, several factors are added at once in
Model 5 only for illustrative purposes because none of the
added factors boost the explanatory power of the model.
These were added one at a time during initial analyses and had
no individual effects.
2230 JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE AND TECHNOLOGY—December 2007
DOI: 10.1002/asi
 15322890, 2007, 14, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/asi.20684 by Wuhan University, Wiley Online Library on [07/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License


JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE AND TECHNOLOGY—December 2007 2231
DOI: 10.1002/asi
Do Social Differences Matter?
At a high level, this study seeks to understand whether
collaboration propensity is best explained by social differ-
ences between fields or by differences in the nature of work
being undertaken. Model 1, in Table 1, suggests that there
are differences between fields when it comes to collabora-
tion propensity, as the field “dummy” variables are strong
predictors of collaboration propensity. As more factors are
added in subsequent models, however, the predictive power
of the HEP variable is gradually eclipsed. In EE, on the other
hand, the dummy variable remains significant. Thus, there is
mixed support for the epiphenomenal explanation of field
differences in that there does appear to be some aspect of EE
that is negatively related to collaboration propensity and is
not effectively captured by the other variables measured
here2. As operationalized in this study, however, social dif-
ferences are measured primarily by focusing on scientific
competition and ease of collective credit attribution.
Scientific Competition
Hypothesis 1 posits a negative relationship between col-
laboration propensity and the perceived level of scientific
competition. This hypothesis was not supported by these
data, as shown in Table 1. The addition of Scientific Competi-
tion in Model 5, along with other factors that had no predictive
power, added no explanatory power over Model 4, and the
standardized /H9252coefficient for this variable was .02 and not
statistically significant. Additional attempts to explore this
surprising result by using individual scale items in place of
the aggregate score yielded no meaningful results.
To be sure, this result should not be interpreted as sug-
gesting that scientific competition does not vary or is not an
important issue in the fields that were studied. Rather, it means
that there is little apparent relationship between concerns
about scientific competition and collaboration propensity. The
qualitative data suggest two reasons for this. 
First, concerns about being the first to make a discovery
do not so much reduce the desire of researchers to collaborate
as much as they constrain the set of possible collaborators to
a few known and trusted colleagues. As one researcher said,
“Collaborations are investigator initiated. And investigators
aren’t going to collaborate with people they think are going
to stab them in the back (N2).”
This was a common feeling among informants and sug-
gests an important distinction between collaboration and the
public release of results and data that were explored in prior
studies. Specifically, the freedom to choose collaborators
and decide when and to whom results will be released means
that concerns about secrecy have more impact on decisions
about whom to collaborate with than whether to collaborate
at all. 
Second, several researchers had strategies for balancing
individual competitive pressures with the need to collaborate.
In one example, two postdoctoral researchers in neuroscience
TABLE 1. Nested linear regression models predicting collaboration propensity ( N /H11005267).
12 3 4 5
Covariates & control
Physics .13* /H11002.03 .02 –.07 /H11002.08
EE /H11002.29*** /H11002.25*** /H11002.26*** –.26*** /H11002.27***
Field tenure /H11002.07 /H11002.03 /H11002.04 /H11002.01 /H11002.02
Coauthor (last 5 yrs) /H11002.01 /H11002.05 /H11002.09* /H11002.06 /H11002.06
Remote coauthor (last 5 yrs) .05 .02 .05 .07 .06
Successful working together .14** .14** .13* .10** .10**
Successful results .08 .09* .08 .05 .05
Network tool usage .17** .11* .07 .10* .11**
Work-Related attributes
Resource concentration .40*** .36*** .25*** .25***
Agreement on quality .26*** .15*** .14***
Need for and availability of .37*** .37***
help
Focus .01
Social factors
Standardized credit attribution .03
Scientific competition /H11002.02
Adjusted R2 .27*** .37*** .43*** .53*** .53***
R2 change 
F Score 13.05*** 42.00*** 29.25*** 57.08*** .22
* p /H11088.1, ** p /H11021 .05, *** p /H11021 .01.
2Note that only EE and HEP are included in Table 1 because “field” is a
qualitative variable with multiple classes (where C /H11005the number of classes),
which Neter, et al. (1996) indicate should be represented by C–1 indicator
variables. In this case, neuroscience is treated as a “reference” variable, indi-
cated by the case where HEP and EE are both set to zero. The effects for EE
and HEP technically indicate the extent to which they differ from neuro-
science. Thus, we can say that EE is consistently lower in collaboration
propensity than neuroscience, while neuroscience and HEP are no different
from each other, after controlling for all other factors.
 15322890, 2007, 14, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/asi.20684 by Wuhan University, Wiley Online Library on [07/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License


2232 JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE AND TECHNOLOGY—December 2007
DOI: 10.1002/asi
reported involvement in collaborative endeavors involving
rare human brain specimens but at the same time maintain-
ing an independent line of smaller–scale research, often
involving rodent specimens, to guarantee some single–
author publications:
The people that I’ve been exposed to with this [large collab-
oration] are all very well known, highly respected people
that have been in the field for a long time and so there’s a lot
to learn from these people. But at the same time I recognize
the importance of independence in this field.  You’re always
pushed to be an independent researcher which is, of course,
the role of the smaller project that I work on (N10).
Ease of Collective Credit Attribution
Hypothesis 2 predicts a positive relationship between col-
laboration propensity and the perceived ease of collective
credit attribution. The data did not support this hypothesis.
As can be seen in Table 1, the addition of this variable in
Model 5 added no explanatory power over Model 4, and the
standardized /H9252coefficient was .03 and not statistically sig-
nificant. Table 2 shows additionally that there does not ap-
pear to be a bivariate relationship between these variables.  
Many interview subjects described situations in which
clearly defined standards for credit attribution simplified the
construction of author lists on articles but led to confusion
and ambiguity when the terms of the collaboration where
changed or where individual contributions needed to be
assessed by outsiders. 
HEP has a longstanding tradition of including all collabo-
rators as authors on any article published by any member of
that collaboration (Galison, 1997). In the most recent genera-
tion of completed experiments, this meant author lists that
included hundreds of names. The LHC experiment author lists
will include thousands of authors. On one hand, this inclusive
tradition is a strong point of pride for many researchers in the
field. It guarantees some degree of formal credit for everybody
who contributes to a project—not just for those who happen to
do the final analysis that leads to a breakthrough result. 
On the other hand, though, there is an important side effect
of such inclusive practices that influences people’s willing-
ness and ability to participate in collaborations. In a landscape
in which the competition for jobs and promotions is fierce, the
use of first– and single–authorships as a measure of individ-
ual accomplishment is essentially impossible in HEP. Thus,
the value of authorship as a mark of distinction is weakened.
Researchers report relying much more heavily on informal
word of mouth reports about colleagues and formal letters
of recommendation. Thus, even though it is clear how
one will receive formal credit in the conventional sense for
contributions to collaborative HEP projects, it is not clear at
all how much actual reward will accrue. (See Birnholtz,
2006 for a more detailed discussion of this issue.)
In neuroscience, standards for authorship are much less
formalized but no less a source of tension and confusion.
Where there are multiple authors on an article in this field,
it is most valuable to be the first or last author. The first
author is generally the researcher who did the bulk of the
work on the project and the last author is the “senior author,”
usually the principal investigator on the grant and the labo-
ratory leader. This distinction works well for small groups
but begins to break down with scale. In large or multisite
collaborations, it may be less clear who should be listed at
which position in the author list, and it is not clear what it
means to be one of the “middle” authors.
If you were the last author, the fourteenth author, the senior
author then it really doesn’t matter for you whether there are
ten other people on it or four other people. But if you’re one
of those fourteen other people or one of three other people, it
makes a difference (N5).
Some researchers described coping with this ambiguity by
agreeing at the start of the project how authorships will be
assigned. This too can cause confusion, however, as the com-
position of the collaboration and the nature of individual con-
tributions can change with time. One researcher described a
case, for example, where she wrote down clear rules in
advance with a collaborator but a third party became involved
late in the project. They then had to rewrite all of the rules,
which she described as a difficult and awkward process.
Indeed, the overall point here is that clear rules for credit attri-
bution may simplify the assignment of authorships in the short
term but can cause ambiguity and confusion with time.
TABLE 2. Bivariate correlations for variables of interest ( N /H11005267).
12 3 4 5 6 7 8
Credit Agree 
Coll. prop. attribution Sci. comp. Focus Res. conc. on qual Help Net. tools
1 1 .07 .06 .19*** .53*** .33*** .57*** .40***
21 /H11002.08 .21*** /H11002.07 .17*** .11* /H11002.07
31 /H11002.04 .07 .14** .09 .11*
4 1 .23*** .26*** .15** .19***
5 1 .14** .38*** .42***
6 1 .34*** .15**
7 1 .18**
8 1
*p /H11088.1, ** p /H11021.05, *** p /H11021.01.
 15322890, 2007, 14, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/asi.20684 by Wuhan University, Wiley Online Library on [07/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License


JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE AND TECHNOLOGY—December 2007 2233
DOI: 10.1002/asi
Considering the Nature of the Work
Hypotheses 3–7 focus on the nature of work carried out by
individual researchers. As is indicated below and in Table 1,
these factors were much more useful in predicting collabora-
tion propensity than the variables in the previous section.
Focus
Hypothesis 3 predicts a positive relationship between the
perceived degree of focus in a field and collaboration
propensity. These data provide mixed support for this
hypothesis. As can be seen in Table 2, there is a moderate
positive correlation between these two variables that is sta-
tistically significant, r /H11005.19, p /H11021.01. Table 1, however,
shows that adding Focus in Model 5 provides no additional
explanatory power over Model 4 and that the standardized /H9252
coefficient is not statistically significant. 
Interview informants suggested two primary reasons for
this. First, the physical scale of research can play an impor-
tant role in determining how much focus is necessary in a
field. In HEP, for example, the scale of the experimental
apparatus necessary to do work essentially mandates a shared
set of research questions, goals, and methods for reaching
those goals. This consensus is constrained and shaped by
issues deemed important by the theoretical physics commu-
nity, the state of detector technology, available funding for
construction and development, and the complex political ag-
gregation process that takes place as new experiments form.
Collaboration is essential if any work is to take place at all,
and there is only a scant handful of experiments in progress at
any one time. Note, though, that this relationship between
focus and collaboration propensity is not purely causal.
Rather, scale is playing a role in driving the need for consen-
sus and focus in the first place.
It is also important to note that HEP is an exceptional case.
Most fields have far more concurrent experiments and put
forth far less effort in reaching or even considering field–wise
consensus with regard to important research problems and
methods. As a result of this, focus can occur at multiple lev-
els and impact collaboration propensity differently at these
levels.
In the EE community, for example, the field is reasonably
rigid in dividing itself into “structural” and “geotechnical”
research communities. Within these subcommunities, it is
far easier to find agreement on research questions and
methods, but even here there are important methodological
differences that could impede collaboration. Some geotech-
nical researchers, for example, primarily use centrifuges in
their experimental work, while others do primarily field work
or use large soil boxes that sit atop hydraulically actuated
shaking tables. There was, however, a great deal of evidence
that such differences in approach are generally viewed as
complementary in the overall community. This stands in
contrast to experiences described in the neuroscience
community where there appeared to be more dissent about
methods and definitions and less respect for alternative
approaches. For example, one subject noted,
Sure, you can get 25 people together in a room and they’ll tell
you, for example, that neural coding is a critical question. But
if you actually sit down with each of those 25 people individ-
ually and ask them what they mean by “neural coding”
they’ll come up with 25 different answers (N1). 
Another subject indicated with respect to his research
methods that there is a large group of researchers “out there
who think this is just crazy, that we really don’t know enough
to knock out or mutate genes and they think we’re never
going to learn anything” (N12). At the same time, however,
the field was generally acknowledged to be large enough that
researchers can find collaborators and simply avoid col-
leagues who do not respect their methods. 
For the present discussion of focus and collaboration
propensity, what all of this means is that focus is compli-
cated and subtle—it can exist at multiple levels and relates to
scale in ways that may not have been fully captured by the
survey instrument used here. More study is needed to better
understand this relationship.
Resource Concentration
Hypothesis 4 predicts a positive relationship between the
perceived level of resource concentration and collaboration
propensity. As can be seen in Table 1, Hypothesis 4 was
strongly supported by these results. Adding resource con-
centration in Model 2 boosts the explanatory power of the
model by a statistically significant difference, F (1, 256) /H11005
42.00, p /H11021.001. Moreover, the standardized /H9252coefficient is
consistently positive and statistically significant in all mod-
els where the variable is present. Thus, there does appear 
to be a positive relationship between the perceived level of
resource concentration and collaboration propensity. The
need for scarce resources of varying sorts appears to push
researchers to work together in accumulating and accessing
these resources to accomplish their goals.
In HEP and EE, the scarce resources are mostly experi-
mental apparatus.  In both fields, a scarcity of funding and
experimental apparatus motivates researchers to work
together. In HEP, this occurs through the pooling of research
resources to construct a detector at a centralized site to
which all of the contributors will then have access. This
sharing is governed by a document called the “Memoran-
dum of Understanding” that specifies contributions and is
signed by a representative from each participating institute.
In the EE community, experimental apparatus are generally
located in a specific university laboratory and “owned” 
by that laboratory. The usage of these resources by “out-
siders” or researchers based at other sites has traditionally
required collaboration with an “insider” who has access to
the equipment. 
In neuroscience, much research is traditional bench science
that can be done in individual laboratories. Here, the resources
that must be concentrated appear to be mostly specific types of
specimens that are difficult to accumulate, such as human
brains, and expertise in different areas of the brain or different
 15322890, 2007, 14, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/asi.20684 by Wuhan University, Wiley Online Library on [07/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License


2234 JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE AND TECHNOLOGY—December 2007
DOI: 10.1002/asi
research methods. Several subjects, for example, reported
specializing in very specific activities, such as one subject
who indicated that he has specialized in a specific method of
statistically processing functional magnetic resonance imag-
ing (fMRI) images. He is entirely dependent on his col-
leagues, mostly at other institutions, to collect fMRI images
and perform preliminary analyses on these. Many researchers
indicated that high quality data sets are extremely costly to
generate and analyze. Collaboration enables these resources
to be pooled and maximally exploited.
Agreement on Quality
Hypothesis 5 states that there should be a positive rela-
tionship between collaboration propensity and the perceived
level of agreement on what constitutes quality research. As
can be seen in Table 1, Hypothesis 5 was supported by these
results. Adding this variable in Model 3 boosts the explana-
tory power of the model by a statistically significant margin,
F(1, 255) /H1100529.25, p /H11021.001. In addition, the standardized /H9252
coefficient for this variable is consistently positive and statis-
tically significant in all models where the variable is present.
There does appear to be a positive relationship between
collaboration propensity and the perceived agreement on
quality. There seem to be two reasons for this: (a) in some
cases, agreement is by design and supports collaboration and
(b) agreement on quality allows researchers to find each
other more effectively.
In HEP, there must be agreement both within and
between the two large LHC collaborations that each experi-
ment will do high quality work. Within the collaboration,
this is critical in that all participating researchers must be
willing to sign their names to an article announcing results in
the end. When individuals fail to agree on the sufficiency of
a result, subjects described systems of extensive discussion
and peer review that are enacted to move the entire collabo-
ration toward consensus. There are also procedures in place
to ensure that no work is published by the collaboration that
has not been carefully reviewed and “blessed” by a desig-
nated committee. 
Second, there is evidence to suggest that agreement on
what constitutes quality research makes researchers more
aware of what their colleagues are doing and therefore bet-
ter able to find collaborators. Where there is agreement,
fields are less fragmented. Researchers read the same jour-
nals, attend the same conferences, and visit each other’s
laboratories to give talks. In some ways, this is similar to
Crane’s (1972) notion of the invisible college. This also
provides a framework for evaluating the work of others.
In neuroscience, for example, interview data suggests
widespread agreement that the top journals are Science,
Nature, and Neuron. 
Need for and Availability of Help
Hypothesis 6 predicts a positive relationship between col-
laboration propensity and the need for and availability of
help. As can be seen in Table 1, Hypothesis 6 was supported
by these results. Adding this variable in Model 4 boosts the
explanatory power of the model by a statistically significant
margin, F (1, 254) /H1100557.08, p /H11021 .001. Additionally, the stan-
dardized /H9252coefficient is consistently positive and statisti-
cally significant.
Generally speaking, this relationship appears to result
from the proximity of researchers to each other in laboratory
spaces. In the case of HEP, this effect is arguably less impor-
tant causally; however, it correlates strongly in that
researchers frequently see each other in the corridors and
cafes at CERN but are mostly already collaborating on a
large scale. In the other fields, though, proximity to noncol-
laborators who do similar work does appear to have an
influence on the amount of help seeking that takes place and
on the volume of discussions that take place that may lead to
eventual collaborations.
The Role of Demographic and Control Variables
Internet–based collaboration tool usage. As can be seen
in Table 1, frequent use of Internet–based collaboration tools
appeared to have a positive relationship with collaboration
propensity, even after controlling for all other factors. On the
one hand, this is not surprising. Researchers who make fre-
quent use of collaboration tools likely have collaborators to
talk to and, therefore, have a higher propensity to collabo-
rate. There is also evidence from a limited number of studies
(Cohen, 1996; Walsh & Maloney, 2002) suggesting that the
usage of computer–mediated communication (CMC) tools
in research work may correlate with increased scientific pro-
ductivity. The finding here serves to reinforce this prior
work, but it still does not address the fundamental question
of whether this relationship is causal and which way the
causal arrow points. As this was not a core focus of this
study, these data offer little to explain this but do point to the
need for additional investigation. 
Individual collaboration experience. Respondents were
asked whether they had participated in collaborative research,
as indicated by publishing one or more articles with coauthors
within the past 5 years, and whether they had done so with
remote coauthors. As can be seen in Model 1 of Table 1, these
variables do not significantly predict collaboration propensity,
though having a coauthor does have a slightly negative rela-
tionship with collaboration propensity in Model 3 that is sta-
tistically significant. This result is unlikely to be meaningful,
however, given its brief appearance and weakness.
Respondents were also asked about the success of recent
collaboration experience, both in terms of working together
well and the results that were achieved. Interestingly, there is
a consistently positive and statistically significant relation-
ship between working together effectively and collaboration
propensity. The same relationship is not present for success
in achieving good results. This combination of results sug-
gests that collaboration propensity depends on social rela-
tionships and the ability to work effectively together, and not
just on the quality of results or having collaborated before.
 15322890, 2007, 14, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/asi.20684 by Wuhan University, Wiley Online Library on [07/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License


JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE AND TECHNOLOGY—December 2007 2235
DOI: 10.1002/asi
Field Tenure
Bozeman and Corley’s (2004) results suggest that tenure
might impact collaboration propensity, depending on the
“collaboration strategy” being used. On the one hand, this
relationship might be positive because tenured researchers
are arguably less concerned about augmenting their individ-
ual reputations. At the same time, though, it could also be
negative in areas where traditional approaches to research
questions call for a single–investigator model. As can be
seen in Table 1, field tenure was not a significant predictor of
collaboration propensity. 
Research Field
It was noted above that the earthquake engineering
dummy variable remained a significant predictor of collabo-
ration propensity even after additional factors were added to
the OLS models. It should also be noted that interaction
effects between the field terms and each of the independent
factors were tested but not found to be statistically signifi-
cant or to contribute to the power of the models3. 
Discussion and Conclusion
This study began with a desire to explain observed differ-
ences in collaboration propensity. While prior studies pro-
vide strong evidence that such differences exist (Knorr
Cetina, 1999; Melin, 2000), it was not clear whether these
differences were best explained by social differences among
researchers, or by the nature of the work being carried out.
The data presented here suggest preliminarily that, in the
fields studied, variance in attitudes toward collaboration is
better explained by the nature of work being conducted than
by the perceived individual versus collectivist orientation of
a field. These results raise a number of issues with implica-
tions for how we think about and measure factors related to
collaboration and collaboration propensity. 
Theoretical Implications
From a theoretical standpoint, this study highlights in
several respects the complexity of the lens with which
collaboration propensity must be approached. In the first
place, this study attempted to separate the social roots of col-
laboration from aspects of scientific work that may increase
or decrease collaboration propensity. While this yielded
some potentially useful results that are discussed below, the
combined quantitative and qualitative data also have two
implications for how we might measure the social aspects of
collaboration. 
First, we must consider the difficulties in identifying and
operationalizing constructs that are reliable indicators of
social differences between fields of research. While the
quantitative data show few direct relationships between col-
laboration propensity and the social factors measured here,
the qualitative data do suggest that there are some subtle
and complex social influences on collaboration. Interview
respondents reported being conscious of scientific competi-
tion and authorship issues in selecting collaborators, for
example, but these factors served more to constrain their
choice of collaborators than to influence overall collabora-
tion propensity. This is further supported by the quantitative
relationship between collaboration propensity and success
with regard to working with colleagues in prior collabora-
tions, and it suggests that there may be additional value
in considering the interpersonal aspects of collaboration
propensity in that researchers seemed much more likely to
collaborate with some colleagues than with others. In other
words, the real issue is not just generic collaboration
propensity but also propensity to collaborate with a particu-
lar individual on a project.
These results also point to the likelihood of interplay
between social attributes of research fields and the nature
of the work in which researchers are engaged. A history of
resource concentration and shared facilities in HEP, for
example, seems to play a strong role in the collective orien-
tation of the field that is evidenced in its attribution prac-
tices and focus on collective entities. At the same time,
though, this relationship is not inevitable. A history of
scarce, shared instruments in astronomy has not had the
same effect in that astronomers tend to work on a wide
range of problems and collaborate in smaller groups 
(McCray, 2000). Much more research is needed to better
understand these relationships, but they do clearly seem to
exist and are worthy of study.
Second, despite the problems in isolating and measuring
the factors studied here, these results do highlight a general
need to enhance our understanding of the conduct of science
by looking more closely at the attributes of scientific work
that impact more broadly observed phenomena, such as col-
laboration propensity. While field–level distinctions clearly
exist, these data suggest preliminarily that there may be sig-
nificant value in looking more carefully at research work at
the micro level. In other words, we may be able to advance
our understanding of collaboration more quickly if we
focus our investigations on researchers with similar styles of
work rather than use fields as boundaries.
Such a shift provides the additional benefits of allowing
consideration of the work attributes studied here more care-
fully and in providing a forum for identifying and isolating
components of research disciplines that may be important to
collaboration but that were not specifically measured here.
In some ways, this general need echoes claims previously
made by Vaughan (1999) and Barley (1996) who call for a
more careful consideration of the organizational setting of
knowledge creation and work. By moving from high–level
studies of science to more detailed and systematic studies
3 These interaction effects were tested for using methods suggested by
Jaccard, Turrisi and Wan (1990). The product terms of each field dummy
variable and each predictor variable were regressed on collaboration
propensity, but the model yielded no statistically significant increase in pre-
dictive power, F (16, 233) /H110051.38, p /H11005.15.
 15322890, 2007, 14, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/asi.20684 by Wuhan University, Wiley Online Library on [07/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License


2236 JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE AND TECHNOLOGY—December 2007
DOI: 10.1002/asi
of scientific work, we can draw on an extensive literature of
individual motivations and group or organizational behavior
and information processing. Such an approach, particularly
as collaboration and the development of collaboration tools
become increasingly important will arguably allow for a
more valuable and nuanced understanding of collaboration.
Practical Implications
From a policy standpoint, the lack of explanatory power
for certain attributes of the individual versus collectivist ori-
entation of the research environment provides some indica-
tion that when considering collaboration tools and their
adoption, certain concerns about scientific competition as an
impediment may be unwarranted. Indeed, a focus on sup-
porting existing collaborative groups, respect for privacy
and allowing them to grow and evolve may be enough to
overcome perceived potential cultural barriers. Researchers
to whom collaboration is valuable may overcome these bar-
riers on their own, and tools should be made readily avail-
able to these researchers as well.
Additionally, these results provide a strong suggestion
that there were work conditions under which attitudes to-
ward collaboration were more positive. In particular, this
was true in areas where resources were highly concentrated
(e.g., at CERN in the extreme case), where there was strong
agreement on what constitutes high–quality research, and
where colleagues were available and consulted when help
was needed. While it is not possible to attribute causality
based on these data, this does indicate that these are circum-
stances under which collaboration may be more likely to
occur. This reinforces prior suggestions, for example, that
much collaboration begins with informal interactions
between those who are proximate (Allen, 1977; Kraut et al.,
1990). Thus, there continues to be likely value in providing
facilities, both physical and online, to support informal in-
teraction that may lead to collaboration.
At the same time, this is not to suggest that merely
concentrating the resources in a field will result in increased
collaboration in that field. This sometimes appeared to occur
in the fields studied here because there were no or few viable
alternatives. It is difficult, for example, to break away from
collaboration as a high–energy physicist and do single–
investigator research because specialized equipment is neces-
sary. The same is not true, for example, in the social sciences,
where it is much easier to do research without specialized
equipment. Thus, concentrating social sciences resources
may not have the same impact on collaboration behavior. 
It is finally important to bear in mind, however, that these
results also provide some limited evidence suggesting that
considering only the nature of the science, and ignoring social
differences, will not always work.  The consistently negative
relationship between being an earthquake engineer and pos-
sessing collaboration propensity indicates that, at least in EE,
there may be some as–yet–unmeasured components of culture
or work that might further inhibit collaboration. 
Limitations and Future Work
Limitations
As suggested above, a major limitation in this work is the
difficulty of effectively measuring social and cultural factors
related to collaboration propensity, and of isolating these
factors from work–related attributes. There are two primary
problems. First, it is difficult to select the correct social fac-
tors and to define these in a way that can be measured. Two
factors were chosen here that have been highlighted in prior
research. However, the fact that there was still a statistically
significant effect for the EE field on collaboration propensity
suggests that there are other social attributes of the
fields studied that influence collaboration propensity but
were not adequately measured here. This suggests that more
research is needed to better operationalize and capture these. 
Second, there is the problem that the work–related attrib-
utes clearly eclipsed the social factors in the quantitative
results, but it is not clear whether this was the purely the
result of explanatory power or, as is more likely, also due in
part to the difficulties of accurately measuring cultural and
social motivators of human behavior. As was mentioned
above, scientific competition and attribution practices do
seem to be related to collaboration propensity, but the rela-
tionships are not always clean or direct. This suggests that
there may be better ways to capture these factors, and to
capture additional factors that may have a more substantial
influence on collaboration propensity, perhaps via more
sophisticated measurement and modeling.
Moreover, there is an implicit assumption here that col-
laboration propensity, as measured here, corresponds to
actual collaboration behavior. While this was not verified in
the context of the present study, this could conceivably be
done by, for example, tracking the number of coauthored
works published by the individuals in the sample for some
future time period. At the same time, however, the variety of
items used to measure collaboration propensity and the
face–to–face contact with many of the respondents in inter-
views do suggest that this construct is at least reasonably
reliable. This is another area for potential future investiga-
tion however.
Additional Directions for Future Research
As mentioned above, this study highlights the complexity
of the lens through which collaboration must be considered
and, in particular, the difficulty of accurately measuring issues
related to collaboration propensity. If we are to engage in the
sort of systematic analysis that will allow us to understand 
the conditions under which collaboration is likely to occur
and succeed, then we must improve our ability to accurately
measure both social and work–related attributes of science
that are related to collaboration. Moreover, it may not always
be possible for survey respondents to fully articulate the mo-
tivations for their behavior or the conditions of work in their
area. Thus, we should consider the development and usage of
 15322890, 2007, 14, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/asi.20684 by Wuhan University, Wiley Online Library on [07/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License


JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE AND TECHNOLOGY—December 2007 2237
DOI: 10.1002/asi
psychometric techniques that allow for inference of attitudes
toward work and collaboration through patterned responses to
novel questionnaire items (similar to personality inventories
used by psychologists).  This sort of work can be combined
with bibliometric and other analyses to derive a much more
nuanced understanding of collaboration propensity.
Acknowledgements
This research was supported in part by the National Sci-
ence Foundation (CMS #0117853), the Horace H. Rackham
School of Graduate Studies at the University of Michigan,
and the Department of Physics at the University of Michi-
gan. I wish to thank Dan Atkins, Matthew Bietz, Michael
Cohen, Paul Edwards, Tom Finholt, Sara Frank Bristow,
Katherine Lawrence, Mike Massimi, Homer Neal, Jason
Owen–Smith and the anonymous reviewers for additional
support and insightful feedback on earlier versions of this
work.
References
Allen, T. J. (1977). Managing the flow of technology. Cambridge, MA: MIT
Press.
Atkins, D. E., Droegemeier, K. K., Feldman, S. I., Garcia-Molina, H.,
Klein, M. L., & Messina, P. (2003). Revolutionizing science and engi-
neering through cyberinfrastructure: Report of the National Science
Foundation blue-ribbon advisory panel on cyberinfrastructure. Washing-
ton, DC: National Science Foundation.
Barley, S. R. (1996). Technicians in the workplace: Ethnographic evidence
for bringing work into organizational studies. Administrative Science
Quarterly, 41(3), 404–441.
Beaver, D. d. (2001). Reflections on scientific collaboration (and its study):
Past, present and future. Scientometrics, 52(3), 365–377.
Birnholtz, J. P. (2006). What does it mean to be an author? The intersection
of credit, contribution and collaboration in science. Journal of the
American Society for Information Science and Technology, 57(13),
1758–1770.
Birnholtz, J. P., & Bietz, M. J. (2003, November 9–12). Data at work:
Supporting sharing in science and engineering. In M. Pendergast,
K. Schmidt, C. Simone & M. Tremaine (Eds.) Proceedings of the Inter-
national ACM SIGGROUP Conference on Supporting Group Work, New
York: ACM Press.
Blumenthal, D., Campbell, E. G., Anderson, M. S., Causino, N., & Louis,
K. S. (1997). Withholding research results in academic life science. Jour-
nal of the American Medical Association, 277(15), 1224–1228.
Bozeman, B., & Corley, E. (2004). Scientists’ collaboration strategies: im-
plications for scientific and technical human capital. Research Policy, 33,
599–616.
Carmines, E. G., & Zeller, R. A. (1979). Reliability and validity assessment.
Thousand Oaks, CA: Sage Publications.
Close, F., Marten, M., & Sutton, C. (2002). The particle odyssey: A journey
to the heart of matter. New York: Oxford University Press.
Cohen, J. (1996). Computer medicated communication and publication pro-
ductivity among faculty. Internet Research, 6(23), 41–63.
Collins, H. M. (1998). The Meaning of data: Open and closed evidential
cultures in the search for gravitational waves. American Journal of Soci-
ology, 104(2), 293–338.
Crane, D. (1972). Invisible colleges. Chicago: University of Chicago Press.
Cronin, B., Shaw, D., & La Barre, K. (2003). A cast of thousands: Coau-
thorship and subauthorship collaboration in the 20th century as mani-
fested in the scholarly journal literature of psychology and philosophy.
Journal of the American Society for Information Science and Technol-
ogy, 54(9), 855–871.
Cummings, J., & Kiesler, S. (2005). Collaborative research across discipli-
nary and institutional boundaries. Social Studies of Science, 35,
703–722.
Engers, M., Gans, J., Grant, S., & King, S. (1999). First author conditions.
Journal of Political Economy, 107(4), 859–883.
Finholt, T. A., & Olson, G. M. (1997). From laboratories to collaboratories:
A new organizational form for scientific collaboration. Psychological
Science, 8(1), 28–36.
Fuchs, S. (1992). The professional quest for truth: A social theory of science
and knowledge. Albany, NY: State University of New York Press.
Galison, P. (1997). Image and logic: A material culture of microphysics.
Chicago, IL: University of Chicago Press.
Galison, P., & Hevly, B. (1992). Big science: The growth of large–scale
research. Stanford, CA: Stanford University Press.
Hagstrom, W. (1965). The scientific community. New York: Basic Books.
Hagstrom, W. O. (1974). Competition in science. American Sociological
Review, 39(1), Feb 1974.
Hara, N., Solomon, P., Kim, S. L., & Sonnenwald, D. H. (2003). An emerging
view of scientific collaboration: Scientists’ perspectives on collaboration
and factors that impact collaboration. Journal of the American Society for
Information Science and Technology, 54(10), 952–965.
Hargens, L. L. (1975). Patterns of scientific research. Washington, DC:
American Sociological Association.
Hesse, B. W., Sproull, L. S., Keisler, S. B., & Walsh, J. P. (1993). Returns to
science: Computer networks in oceanography. Communications of the
ACM, 36(8), 90–101.
Jaccard, J., Turrisi, R., & Wan, C. K. (1990). Interaction effects in multiple
regression. Newbury Park, CA: Sage Publications.
Katz, J. S., & Martin, B. R. (1997). What is research collaboration?
Research Policy, 26, 1–18.
Kennedy, D. (2003, August 8). Multiple authors, multiple problems. Sci-
ence, 301, 733.
Knorr Cetina, K. (1999). Epistemic cultures: How the sciences make
knowledge. Cambridge, MA: Harvard University Press.
Kouzes, R., Myers, & Wulf, W. (1996). Collaboratories: Doing science on
the internet. IEEE Computer, 29(8), 40–46.
Kraut, R., Egido, C., & Galegher, J. (1990). Patterns of contact and com-
munication in scientific research collaborations. In J. Galegher, Kraut,
R.,& Egido, C. (Ed.), Intellectual teamwork. Hillsdale, NJ: Lawrence
Erlbaum Associates.
Laband, D. N., & Tollison, R. D. (2000). Intellectual collaboration. Journal
of Political Economy, 108(3), 632–662.
Landry, R., & Amara, N. (1998). The impact of transaction costs on the in-
stitutional structuration of collaborative academic research. Research
Policy, 27, 901–913.
Malone, T. W., & Crowston, K. (1994). The interdisciplinary study of coor-
dination. ACM Computing Surveys, 24(1), 87–119.
McCray, P. (2000). Large telescopes and the moral economy of recent astron-
omy. social studies of science, 30(5), 685–711.
Melin, G. (2000). Pragmatism and self–organization: Research collabora-
tion on the individual level. Research Policy, 29, 31–40.
Mervis, J. (2005, April 29). Marburger asks social scientists for a helping
hand in interpreting data. Science, 308, 617.
Miles, M. B., & Huberman, A. M. (1994). Qualitative data analysis: An
expanded sourcebook. Thousand Oaks: Sage Publications.
Nentwich, M. (2003). Cyberscience: Research in the Age of the Internet.
Vienna: Austrian Academy of Sciences.
Neter, J., Kutner, M. H., Nachtsheim, C. J., & Wasserman, W. (1996). Applied
linear statistical models. Boston: WCB McGraw–Hill.
Newell, A., & Sproull, R. F. (1982, March 5). Computer networks: Prospects
for scientists. Science, 215(4534), 843–852.
Newman, M. E. J. (2001). Who is the best connected scientist? A study of
scientific coauthorship networks. Physics Review E, 64(4), 16131.
Oliver, A. L. (2004). Biotechnology entrepreneurial scientists and their
collaborations. Research Policy, 33, 583–597.
 15322890, 2007, 14, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/asi.20684 by Wuhan University, Wiley Online Library on [07/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License


2238 JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE AND TECHNOLOGY—December 2007
DOI: 10.1002/asi
Price. (1963). Little science, big science. New York: Columbia University
Press.
Sims, B. (1999). Concrete practices: testing in an earthquake engineering
laboratory. Social Studies of Science, 29(4), 483–518.
Thompson, J. D. (1967). Organizations in action: Social science bases of
administrative theory. New York: McGraw–Hill.
Traweek, S. (1988). Beamtimes and lifetimes: The world of high energy
physicists. Cambridge, MA: Harvard University Press.
Van De Ven, A. H., Delbecq, A. L., & Koenig Jr., R. (1976). Determinants
of coordination modes within organizations. American Sociological
Review, 41(2), 322–338.
Vaughan, D. (1999). The role of the organization in the production
of techno–scientific knowledge. Social Studies of Science, 29(6),
913–943.
Wagner, C. S., & Leydesdorff, L. (2005). Network structure, self–organiza-
tion and the growth of international collaboration in science. Research
Policy, 34, 1608–1618.
Walsh, J. P., & Hong, W. (2003, March 20). Secrecy is increasing in step
with competition. Nature, 422, 801–802.
Walsh, J. P., & Maloney, N. G. (2002). Computer network use, collabora-
tion structures, and productivity. In P. Hinds & S. Kiesler (Eds.), Distrib-
uted Work. Cambridge, MA: MIT Press.
Whitley, R. (2000). The intellectual and social organization of the sciences.
Oxford: Oxford University Press.
Who’d want to work in a team? (2003, August 14). Nature, 424, 1.
Zimmerman, A. (2003). Data sharing and secondary use of scientific data:
experiences of ecologists. Unpublished doctoral dissertation, School of
Information, University of Michigan, Ann Arbor.
Appendix A
Questionnaire Items and Factor Component Scores
Factor analysis
Variable Item wording component scores*
Agreement on quality When I assess the work of my peers, I use the same standards that they use in .53
assessing my work
Agreement on quality When I assess the merits of a peer’s research, my assessment is generally in agreement .64
with my peers.
Agreement on quality When my work is reviewed by my peers, I generally agree with their assessment. .56
Agreement on quality There is a clear hierarchy of journals in my field, with leaders that are generally .65
agreed upon throughout the field
Agreement on quality There is a clear hierarchy of universities in my field, with leaders that are generally .49
agreed upon by most researchers
Availability of and need for help I frequently come across specific, difficult problems in my work that I do not know .61
how to solve alone.
Availability of and need for help In doing my day-to-day research work, I use a standard set of methods that could also .54
be applied to other problems or tasks.
Availability of and need for help When I encounter a difficult problem in my work, I seek the advice of a colleague or mentor .69
Availability of and need for help Most other researchers in my field use techniques or methods similar to the ones .61
that I use
Collaboration propensity Collaboration with other researchers would benefit my career. .54
Collaboration propensity Other researchers in my field who do collaborative work are successful. .62
Collaboration propensity I plan to engage in collaborative research in the future. .74
Collaboration propensity Collaboration is necessary in my field .69
Collaboration propensity Collaboration is useful in solving problems that are of interest to me .81
Credit attribution practices When I participate in a research project, it is clear at the start of the project how I will .81
receive credit for my contribution to the work (i.e. via an authorship on a publication, etc.)
Credit attribution practices When I publish an academic paper, it is easy to determine whom to include as coauthors .81
on the work
Focus The methods I use in my research are the only methods used for legitimate research in my .64
field
Focus There are methods used by some prominent researchers in my field that I do not believe .51
yield valid results even when they are used correctly
Focus There is widespread agreement in my field about what the important research questions are .73
Resource concentration Doing cutting edge research in my field requires access to rare and expensive equipment .75
Resource concentration Producing quality research in my field requires access to an amount of funds that it might .75
be difficult for a single investigator to secure
Scientific competition I feel safe in discussing my current work with other persons doing similar work (other .48
than my collaborators)
Scientific competition I am concerned that the results of my current research might be anticipated or “scooped” .63
by other scientists working on similar problems.
Scientific competition The competition for prizes or widespread recognition in my field is intense .30
Scientific competition In the past 5 years, the results of my research have been anticipated or “scooped” by .45
other scientists working on similar problems
Note. * A separate factor analysis was run for each variable.  In all cases, principal components analysis was the extraction method and a single
component was extracted.
 15322890, 2007, 14, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/asi.20684 by Wuhan University, Wiley Online Library on [07/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License


JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE AND TECHNOLOGY—December 2007 2239
DOI: 10.1002/asi
Appendix B 
Descriptive Statistics
Descriptive statistics by field (N /H11005267).
Physics (N /H11005103) EE ( N /H1100591) Neuro ( N /H1100573) Total
Variable Min Max Mean SD Min Max Mean SD Min Max Mean SD Mean SD
Collaboration 16.00 25.00 22.06 2.07 14.00 25.00 19.34 2.33 15.00 25.00 21.10 2.44 20.87 2.54
propensity
Scientific competition 5.00 17.00 11.64 2.49 6.00 19.00 11.76 2.42 4.00 18.00 11.84 2.95 11.73 2.60
Credit attribution 2.00 10.00 7.47 1.58 4.00 10.00 7.69 1.28 3.00 10.00 7.19 1.52 7.47 1.47
Focus 4.00 13.00 8.87 1.58 4.00 11.00 7.65 1.49 4.00 10.00 7.21 1.33 8.00 1.64
Need for and 11.00 20.00 16.72 1.72 9.00 20.00 15.63 2.01 10.00 20.00 15.78 1.94 16.09 1.94
availability of help
Agreement on 14.00 23.00 19.00 1.81 13.00 25.00 18.85 2.18 12.00 24.00 18.86 2.15 18.91 2.03
quality
Resource 4.00 10.00 9.57 .98 3.00 10.00 7.30 1.80 3.00 10.00 7.86 1.77 8.33 1.83
concentration
Note. All variables were measured using 5-point Likert scales, but different numbers of item scores were summed to create these constructs. Thus, 
the value ranges for these variables differ. Overall maximum and minimum values are not provided in the ‘total’ column because these can be easily
determined by looking for the smallest or largest value in each row.
 15322890, 2007, 14, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/asi.20684 by Wuhan University, Wiley Online Library on [07/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License