Network Science 1 (1): 1–15, 2013.


c Cambridge University Press 2013 1
doi:10.1017/nws.2013.2

EDITORIAL
What is network science?
ULRIK BRANDES
Department of Computer and Information Science, University of Konstanz, Germany
(e-mail:)[email protected])

GARRY ROBINS
Melbourne School of Psychological Sciences, University of Melbourne, Australia
(e-mail:)[email protected])

ANN McCRANIE
Department of Sociology, Indiana University Bloomington, USA
(e-mail:)[email protected])

STANLEY WASSERMAN
Departments of Psychological and Brain Science and Statistics, Indiana University Bloomington, USA
(e-mail:)[email protected])

Abstract
This is the beginning of Network Science. The journal has been created because network
science is exploding. As is typical for a field in formation, the discussions about its scope,
contents, and foundations are intense. On these first few pages of the first issue of our new
journal, we would like to share our own vision of the emerging science of networks.

1 Is there a science of networks?

Whether there is or is not a science of networks, and what should be the focus of
such a science, may appear to be academic questions. We use this prominent place
to argue the following three statements:
• These questions are more relevant than they seem at first glance.
• The answers are less obvious than have been suggested.
• The future of network science is bright.
Of course, the editors of a new scientific journal are going to predict that the future
of their science is bright. Cambridge University Press is betting that it is. We do
also think, however, that the field is still in formation and that its essential shaping
is taking place right now. The development and steering of network science into a
fertile and beneficial direction form our primary motivations for creating this journal
and for assuming active roles in its leadership.
Network science, the field, permeates a wide range of traditional disciplines,
and Network Science, the journal, will welcome contributions from all of them.
In addition, we will indeed publish foundational research on theory, principles,
philosophy, and mathematics of networks.

Downloaded from https://www.cambridge.org/core. Pontificia universidad catolica de valparaiso, on 22 Mar 2018 at 18:28:55, subject to the Cambridge Core
terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/nws.2013.2
 2 U. Brandes et al.

While there is much overlap in disciplinary network research, the inherent

disciplinary boundaries still tend to create silos of different interests, methods,
and goals. If network science is to be one science, rather than separate and scattered
research communities, or a set of tools that researchers use to analyze networks, the
silos need to be dismantled while at the same time recognizing existing disciplinary
practices and values.
We write this editorial to help establish common grounds for our journal and
its field—grounds which should allow Network Science to excel above and beyond
disciplinary boundaries. We envision our commons as much wider than some current
interpretations of the term network science. We will therefore try to delineate the
uniting elements rather precisely on the next few pages.
Our major statement is that we view network science as the study of the collection,
management, analysis, interpretation, and presentation of relational data. But first, a
few remarks on our perceptions of the current state of the field.
The claim that “networks are everywhere” has become almost routine. Frequently
mentioned examples of “everywhere networks” include the Internet and other
infrastructure networks, social, political and economic networks, scientometric and
text-representational networks, as well as food webs and molecular-level biological
networks. And there is a host of other, less commonly mentioned networks in many
more research areas.
Networks and hence the network paradigm have become scientifically relevant
across disciplinary boundaries. But many have asked: Is the network paradigm
nothing more than an in vogue buzz phrase? Clearly, a science of networks requires
a scientific commonality that creates a foundation for these different research streams.
We must ask: Is there such a unity? And if there is, does this unified, shared element
really contribute something relevant to science, or does it only appear to do so by
virtue of the fact that “networks are hot?”
These questions are important to answer because out of this acclaimed relevance
grows a promise— a promise of a new scientific discipline.
When we are surrounded by networks, with many of them arising from topics of
the highest societal relevance, we certainly feel the need to understand them. Some
researchers, at their most enthusiastic, have gone as far as proclaiming that network
science is not just helpful, but essential. In a televised documentary (on Australian
Broadcasting Corporation (ABC), Australia, in 2008), Duncan Watts said it very
bluntly:
“Networks are important because if we don’t understand networks, we can’t
understand how markets function, organizations solve problems, or how
societies change.”

In a mathematical sense, this statement implies the necessity of network science.

Given the many reports and commentaries flirting with the idea that network science
may hold the key to just about every important problem, one is tempted to think
that it could even be sufficient (again, in a mathematical sense).
While we are skeptical about the singularity and uniqueness of network science, we
do acknowledge its potential relevance. It is quite amazing how many researchers,
from different disciplines, have stated in survey and position papers that their
research and discipline are being revolutionized by the increased use and importance

Downloaded from https://www.cambridge.org/core. Pontificia universidad catolica de valparaiso, on 22 Mar 2018 at 18:28:55, subject to the Cambridge Core
terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/nws.2013.2
 Editorial 3

of the network paradigms. Political Science is a good example—network science has

had a tremendous impact on this field, just in the last five years.
If anything, network science is a revolution a long time in the making. Despite
frequent claims by some, network science did not suddenly appear when it was
realized in the mid-1990s that networks could be models of complex systems. Such
a limited definition of network science is simply inappropriate—it is important to
recognize the many scientific antecedents of what we do. Network approaches have
developed in many areas over the past two decades (physics, biology, economics, for
example) because a relational perspective clearly added relevance to the discipline.
The roots of network science are particularly strong in social psychology, sociology,
and anthropology, which has led to another misperception, namely that network
science is the application of network analysis in disciplines other than the social
and behavioral sciences. Sometimes the phrase social network analysis (“SNA”) is
used to label everything that is network-related, even when the network aspects of
the work are clearly not analysis at all (e.g., network theory). Acronyms may be
appealing, but if all we do is “analysis,” we simply will not be able to create a real
network science, fulfilling grand promises.
The network perspective allows us to address deep questions about human,
biological, economic, and other systems that exhibit interdependent organization.
As network researchers, and as editors of this journal, we do commit to the above
promises, despite knowing that many of them will always remain promises.
At this stage of our science, there are too many areas in which we cannot point
to empirical results that convincingly meet the claims being made. The Human
Connectome Project is but one example that is just barely allowing us to study
meaningful network connections in the human brain. Moreover, results of network
studies are seldom tested against alternative conceptualizations—just as alternative
approaches seldom test themselves against a network approach.
We want the science of networks to develop beyond claims and promises. We do
not want to see the field degenerate into a scientific fad, leaving behind discredited
theories and ideas. With this journal, we will try to advance the unity of a scientific,
non-metaphoric perspective, and thus help bring its promises to fruition.
As we shall explain, the single, deepest point that unites us is not a toolbox
of methods but a conceptualization that appears way before there is anything to
analyze at all.

2 Network science and network theory

The above mentioned claim that “networks are everywhere” implies that the many
phenomena jointly referred to as “networks” belong to a common category. For
the diverse phenomena being studied this is only possible by means of abstraction
so that a network and the phenomenon referred to as the network are generally
distinct. The diagram in Figure 1 shows the role of abstraction as a conceptual
prerequisite for representation in data.
Consequently, a network model should be viewed explicitly as yielding a net-
work representation of something. While the model manifests itself in a network
representation, it is obtained via the following two steps:

Downloaded from https://www.cambridge.org/core. Pontificia universidad catolica de valparaiso, on 22 Mar 2018 at 18:28:55, subject to the Cambridge Core
terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/nws.2013.2
 4 U. Brandes et al.

abstraction representation
phenomenon network concept network data
network model

Fig. 1. The elements of network models.

1. A specification of how the phenomenon (in general, i.e., more generally than
this particular instantiation) is abstracted to a network.
2. A specification of how this conceptual network is represented in data (e.g.,
measured or observed).

As representation is usually defined via an isomorphism, i.e., a one-to-one mapping

between structures preserving relations, a phenomenon cannot be represented
directly but needs to be conceptualized first.
Of course, this is by no means an unusual division in science or other areas of
knowledge. Possibly because of the graphic and metaphoric connotations of the
term network, the implications of a preceding abstraction step are often overlooked
or blurred. Sometimes this may be on purpose for terminological convenience.
More often, there appears to be a lack of awareness. We feel, however, that this
distinction is crucially important for serious applications of network science to the
understanding of substantive phenomena as it points to the delicacy of interpreting
the results of network data analysis.
Interpretation essentially reverses the process of abstraction and representation
to get back to the phenomenon so that substantive theory is required to secure
conclusions. Abstraction facilitates generalization and comparison but at the same
time presents an obstacle in this reverse direction, when statements about a network
representation are interpreted in terms of the original phenomenon. The means to
overcome this obstacle are necessarily disciplinary. We therefore consider network
science to be the study of particular kinds of representations; issues relating to
conceptualization, however, form an integral part of the application of network
science to substantive problems.

Claim 1
Network science is the study of network models.

We emphasize that, in our view, network models are unlikely to generalize across
domains. We hence remain open to, but rather sceptical about any Grand Unified
Network Theory (GrUNT) that ignores research contexts. On a related note, we find
it very unfortunate that many network studies are referred to as “social network
analysis” just because the methods applied are commonly used for social network
models. If the network under scrutiny models, say, gene regulation, the term is
hardly appropriate.
Network theory builds on the assumption that a cause, an effect, or an association
between aspects involve something that can be conceptualized as a network. Testing
of hypotheses derived from network theory requires instantiation of the network
model. While the abstraction yields the format in which a phenomenon will be
represented, its actual representation is in terms of data that is typically obtained
via empirical observation.

Downloaded from https://www.cambridge.org/core. Pontificia universidad catolica de valparaiso, on 22 Mar 2018 at 18:28:55, subject to the Cambridge Core
terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/nws.2013.2
 Editorial 5

According to our framework there are actually two aspects to network theory.
On the one hand, network theories can suggest and explicate, for given research
domains, how to abstract phenomena into networks. This includes, for example,
what constitutes an individual entity or a relationship, how to conceptualize the
strength of a tie, etc. In such applied network science, the corresponding theories
are epistemological—network theories bound to specific classes of phenomena. On
the other hand, network theories can deal with formalized aspects of network
representations such as degree distributions, closure, communities, etc., and how
they relate to each other. In such pure network science, the corresponding theories
are mathematical—theories of networks.

Claim 2
There are theories about network representations and network theories about
phenomena: both constitute network theory.

Establishing network theory can be a challenge in disciplines that have a highly

individualized history. But without a theory about how to conceptualize a phe-
nomenon as a network, there is no meaning to a formal theory of network data. We
conclude that networks are not just an add-on to existing approaches—e.g., a means
to add a little more explained variance in a social science research project—but
require new theorizations and different thinking.
A network abstraction involves ontological commitment to a few basic features
that are seen as scientifically relevant to the representation of a phenomenon. The
features of a network abstraction include at least the following: individual elements;
pair-wise relationships between those elements; and a global or macro- patterning
that can be considered as network structure. This basic description may not be
sufficient for all circumstances (e.g., think of longitudinal phenomena) and can be
extended in many different ways, but these are fundamental features if we are to
call the abstraction a network.
For example, a friendship network is a way of abstracting a social phenomenon
into a comparatively simpler and much more general form of relationships between
actors. The actual phenomenon is, of course, much richer: We are abstracting already
quite substantially just by conceiving of the individuals as comparable entities of a
common kind.
By postulating a friendship network in (say) a school classroom of 25 students,
we have taken a theoretical step that is non-trivial. We have supposed that separate
individuals are not an adequate representation, moreover that even separate dyads
are insufficient; rather, that there is a unity within the classroom that makes it
proper to talk of “a” network, not 25 children or 300 dyads.
To conceptualize the classroom in network terms is an implicit (and strong) claim
that connectedness across individual elements is fundamentally important so that
the classroom can be thought of as one “system.” If we accept that ontology,
scientific inference is available at multiple levels: the students, the dyads, and indeed
the network as a whole. Moreover, the inferences at one level cannot be simply
combined (e.g., averaged) to derive inferences at other levels; the networked system
is more than a simple aggregation of its constituent elements—it is patterned, not
summed.

Downloaded from https://www.cambridge.org/core. Pontificia universidad catolica de valparaiso, on 22 Mar 2018 at 18:28:55, subject to the Cambridge Core
terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/nws.2013.2
 6 U. Brandes et al.

So the claim that “networks are everywhere,” if it is meaningful as network science,

is not just a statement that we can see many things in the world in relational terms,
but an implicit theoretical statement that scientific explanation of many phenomena
is aided by abstraction to such a connected, systemic representation. Otherwise, it
is no more than a statement that we can see the world in particular ways: after all,
colors are everywhere, too, but no one to date has thought it scientifically helpful to
understand classroom processes in shades of pink and purple.
The essential point must be that the abstraction into a network is helpful to
scientific inference, permitting knowledge to develop. It need not be the case that
this will always be so. But we need an empirical base to show that the network
representation gives scientific traction.
Claim 3
Network science should be empirical—not exclusively so, but consistently—and its
value assessed against alternative representations.

3 Network data
We have argued that networks are abstractions represented in data, but we have yet
to discriminate them from other conceptualizations. We are now going to do so by
first looking at characteristics of standard types of data to be able to then highlight
the defining features of network data.
The input to data analysis consists of values of variables. Variables are generic
placeholders characterizing the essential features of an abstract concept, thus
allowing to formulate analytical steps generically as well. The instantiating values
are usually obtained via some form of observation such as measurement. Note,
however, that different original phenomena may yield the same representation in
data.
Our definition of what constitutes network data hinges entirely on how the
involved variables are related. It is thus independent of the phenomena being
represented, the ranges of values that can be assumed, and the techniques used to
analyze the data. The significance of the following claim will be established in three
steps below.
Claim 4
What sets network data apart is the incidence structure of its domain?
The third step will also unveil the essential correspondence between the signature
of network variables and the defining interest in network science.
Claim 5
At the heart of network science is dependence, both between and within variables.

3.1 Step 1: Data tables

In data analysis, variables are generally associated with a set of entities (e.g.,
individuals), so that the same attribute is measured for all entities. The entities
form a (sample from a) population and the variables corresponding to the same
attribute are combined into a vector indexed by the entities. Variables for different

Downloaded from https://www.cambridge.org/core. Pontificia universidad catolica de valparaiso, on 22 Mar 2018 at 18:28:55, subject to the Cambridge Core
terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/nws.2013.2
 Editorial 7

A x
a 1
b 4
c 1
d 2
e 3
f 6
(a) standard table: variables in columns indexed with unrelated entities

D x
(a, f ) 1
(d, e) 5
(b, c) 3
(b) dyadic: variables in columns indexed with unrelated pairs of entities

D x x(D) a b c d e f
(a, b) 0 a · 0 1 0 1 0
(a, c) 1 b 0 · 1 2 · ·
(a, d) 0 c 1 1 · 0 2 0
(a, e) 1 d 0 2 0 · 1 4
(a, f ) 0 e 1 · 2 1 · 2
(b, c) 1 f 0 · 0 4 2 ·
(b, d) 2
(c, d) 0
..
.
(c) network: variables in columns indexed with incident pairs of entities, or in matrices

Fig. 2. Data formats distinguished by the structure of the domain.

attributes are conveniently organized in a matrix where each column corresponds

to all variables representing the same attribute and each row corresponds to all
variables associated with the same entity. This is illustrated in Figure 2(a).
A prototypical example would be a set A of individuals with variables gender,
education, and income, where, e.g., a variable incomei represents the gross income
of an individual i ∈ A in Euros per year. The data for one attribute is thus combined
in a vector with index set A. But one alternative representation is that of a mapping
income : A → R, where R is an interval containing possible values. More generally,
the notation x : D → R declares a mapping (representing the variable) x that assigns
to each element in the domain D a value in the range R. The elements of the domain
are units of observation or analysis.

Downloaded from https://www.cambridge.org/core. Pontificia universidad catolica de valparaiso, on 22 Mar 2018 at 18:28:55, subject to the Cambridge Core
terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/nws.2013.2
 8 U. Brandes et al.

In the above example, variables gender, education, and income are of different
types: While all are defined on the common domain A, the range of values they may
assume is different. Even more importantly, these ranges exhibit a different level of
structuring. The range of gender, for instance, is binary on a nominal scale, i.e.,
the only defined relation is an equality predicate. In other words, the comparison of
two values yields either equality or inequality, and this is the only information we
can get out of comparison. For instance, we cannot add or rank values of gender
variables.
Assume now that education refers to the highest degree obtained by an individual.
It may then be valid to compare two values and conclude that one indicates a higher
level of education than the other, and this relation could be transitive. In this case,
the range is ordered and the variable is on an ordinal scale of education. Finally,
we may compare income by amount, but it may also be meaningful to compute
differences and ratios. The range of the variable income can therefore be considered
to be a continuous ratio scale. If, however, 0 is not meaningful for a continuous
range as in, e.g., measuring IQ, it is not appropriate to calculate ratios and the scale
is called interval.
The interesting thing to observe is that a range is usually not just a set of possible
values but a set with additional relations such as an ordering or operations, i.e.,
structured. The structure of a range is crucial to know about because it determines
the kinds of analyses and interpretations that are justified.
While the range of attributes is structured, in much of science, the domain on
which variables are defined is assumed to have no structure, i.e., simply a set. This
may be for good reason. If we are interested in associations between, say, education
and income controlled for age, we actually do not want there to be relations
between individuals that also moderate the association. Much of statistics is in fact
concerned with detecting and eliminating such relations.
This is the single most important difference with network science, where the
domains of at least some variables are explicitly set up to have structure. The
potentially resulting dependencies are not a nuisance but more often than not they
constitute the actual research interest.

3.2 Step 2: Dyadic data

Before introducing the structure of network-variable domains, consider as an
intermediate step dyadic variables, i.e., variables defined on pairs of items. A classic
example of this kind is the study of (populations of) couples. Here variables, such
as duration of marriage, number of common children, etc., are associated with the
couples, whereas further variables, such as age, occupation, etc., are associated with
the individuals that make up the couples.
As illustrated in Figure 2(b), the domain of couple-level variables is therefore
composed of pairs of individuals that cannot be treated as a new, atomic unit because
it is important to maintain individual identities to be able to find between-variable
associations in individual- and couple-level data. Similarly, attributes of couples
cannot be represented in individual-level variables because this means eliminating,
for instance, possible between-variable associations of individuals and marriages
that are moderated by attributes of spouses.

Downloaded from https://www.cambridge.org/core. Pontificia universidad catolica de valparaiso, on 22 Mar 2018 at 18:28:55, subject to the Cambridge Core
terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/nws.2013.2
 Editorial 9

A domain representing couples is structured but only minimally so. The only
relation is a pairing of individuals in dyads. For the statistical analysis to work, it is
usually desirable that these dyads be disjoint and independent. For example, having
several individuals appear in two or more distinct dyads may invalidate findings
about associations between, say, age differences and the number of common children.
In this respect dyadic data is not all that different from standard data tables.

3.3 Step 3: Network data

While the units of observation for network data (the entities with which attributes
are associated) are dyadic as well, it is by design that these dyads are overlapping, i.e.,
they intersect. Two dyads with a common member are called incident. See Figure 2(c)
for an illustration.
If we are interested in, for instance, the effects of friendship among children of
a school class, and we observe friendship ties for each pair of them, then the mere
fact that the same child appears in several dyads creates a potential dependency
between the attributes of those dyads. This is of course not a defect of the study
design. Rather, the patterns found in data on incident dyads are the essence of what
we are after because they may help explain, e.g., which particular children can be
considered crucial for the integration of the class.
Concretely, we can define a network variable as a mapping x : D → R from a
domain of non-disjoint dyads D to a range R. In the common case of interaction
domains D ⊆ A × A, the units of observation are pairs of entities, or actors,
from a set A. A relation x then corresponds to a one-mode network that can be
represented, for instance, as a graph or a square matrix. Note that there may be
so-called structural zeros, i.e., entries (k, l) ∈ D that are not part of the analysis, often
the diagonal entries.
The domain can be more general, though. An affiliation domain D ⊆ A × S relates
actors in A (e.g., directors) with settings in S (e.g., company boards). The two-mode
networks defined on such domains can be represented, for instance, in a rectangular
matrix, a bipartite graph, or a hypergraph. An ensemble of ego networks can then
be considered as a special two-mode network in which each alter in A is affiliated
with exactly one ego in S, i.e., dyads overlap only on the ego side.
Diametral to standard forms of data analysis, dependencies between the elements
of the domain are at the heart of network representations. Of course, we most often
think of two attributes being dependent on, or associated with, each other. For
instance, it is possible that income may be associated with age. In standard terms
that is an association between the distributions of two variables, homogeneous with
respect to individuals. It does not matter whether individual i is rich or old (or
both): rather, if there is an association between the two variables, then wherever i
is located on one of the distributions (age or income) tends to be associated with
i’s location on the other distribution; moreover, this association is in some sense an
average across all such individuals i.
As a consequence that may not immediately be obvious, an incidence-structured
domain leads to interesting questions also about within-variable associations, or
patterned relations, beyond distributional analyses. The first conceptual extension
is that the income of i may be related specifically to the income of j, e.g.,

Downloaded from https://www.cambridge.org/core. Pontificia universidad catolica de valparaiso, on 22 Mar 2018 at 18:28:55, subject to the Cambridge Core
terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/nws.2013.2
 10 U. Brandes et al.

because they are good friends who share information relevant for their salary
negotiations. In other words, rather than being a dependence between two different
types of variables, now we have a dependence within the values for one type
of variable. This is a complex dependence because it cannot be aggregated or
averaged in distributional terms. It corresponds to the kind of dependency ana-
lyzed in spatial statistics (with proximity rather than friendship as the underlying
mechanism).
Yet, network dependence goes further because dependence does not just stop
at actor attribute variables. It may apply within the set of network variables as
well. Any network variable is defined on a domain of pairs of individuals (i.e.,
the dyads), and the incidence structure of the domain captures the potential for
within-variable dependencies. A network tie variable takes a value, often binary,
sometimes valued, indicating whether there is or not a tie between its two individuals.
The crucial point is that the presence of one tie may influence the presence of
another. In other words, ties are not necessarily moderating variables, but there
may be dependencies within the tie variable themselves. While this will appear an
unfamiliar point of view to some, it is merely a statement that networks may be sys-
tematically patterned. Without dependence among ties, there is no emergent network
structure.
In the explicit form of stochastic models, these ideas entered network analysis
from spatial statistics. They are deeply at the heart of network theory, even if seldom
overtly addressed. Entire sets of methodological approaches, such as exponential-
family random graph models, depend on modeling tie dependence appropriately.
With independence among network tie variables, we would be left only with
the simple random networks known as Bernoulli graphs, Erdős-Renyi graphs, or
the G(n, p) model. It should be noted that this view does not require a statistical
perspective; combinatorial invariants of graphs that represent networks are of
interest exactly for the same reason as descriptors of structural features.
Because almost all the networks that we observe bear little resemblance to simple
random graphs, tie dependence is empirically very common. For instance, a familiar
network process is that of preferential attachment, whereby actors “prefer” to be
attached to popular actors so that the rich get richer. The presence of many ties
centered on one popular individual may attract the presence of additional ties to
that same individual.
Dependence among ties is thus the means whereby network structure self-organizes
and evolves, or emerges, but it is not simple. This is why network science is often
referred to as the study of complex networks. It remains a research question to
establish plausible types of tie dependence. Theories or methods that wish away these
dependencies are ignorant of the structure of the domain, and thus contradictory to
a network model.
While the choice of representation is indeed a matter of convenience and hence
relative to any given scenario, we think that the data-oriented perspective of a
structured domain is leaner than, for instance., the common strategy of defining
networks in terms of graphs. Graphs may be one of the most common forms of
representation but they do not make the most distinguishing feature of network data
apparent. The edges display the structure of an observation, not of the conceptual
setup that led to it.

Downloaded from https://www.cambridge.org/core. Pontificia universidad catolica de valparaiso, on 22 Mar 2018 at 18:28:55, subject to the Cambridge Core
terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/nws.2013.2
 Editorial 11

At Network Science, we anticipate to publish work on all kinds of network data,

including ego networks. We also hope to receive contributions on network sampling
techniques, possibly defining the domain on the fly, and even fundamental theoretical
work on structured dyadic domain.
Contributions using other forms of data, simulations, or analytical approaches
that focus on building theory, illustrating practical data concerns, or fleshing out
nuances not recognized in standard representations are welcome as well. Likewise is
rigorous qualitative research which could offer insight into process and meaning that
can be missed otherwise. With all articles, the quality of data and the appropriate
use of it to answer the questions put forth about relational, complex phenomena is
the most important criterion.

3.4 Discussion
Our characterization of network data focuses on the structure of the domain of
variables irrespective of their range. We have built our exposition on this definition
because it carves out the essential distinction from standard data and allows for
a more uniform description of different network types. A further advantage is the
clear distinction between dyads for which no data is available (because they are not
in the domain) and dyads for which the data indicates the absence or nullity of a
relationship.
Statistics is often defined as the study of data, involving anything from its col-
lection, preparation, and management to its exploration, analysis, and presentation.
In this view, our definition of network science delineates a subarea of statistics
concerned with data of a peculiar format. This implies that, like general statistics,
network science is not tied to any particular substantive area. The disciplines for
which area editors have assumed responsibility should therefore not be viewed as
limiting the scope of submissions.
Areas such as machine learning or data mining are distinguished from other areas
of statistics by the tasks addressed. Operations research is distinguished mostly
by the subject matter and its consequences for data and tasks. In contrast, it is
the special data format that causes network science to be markedly different from
other areas in statistics. First, the abstraction to a different category of concepts
introduces a combinatorial structure on the domain of variables. This structure
leads to alternative representations that ask for more combinatorial approaches than
distributions and graph theory in particular. Second, there is an inherent focus on
interdependence and (in contrast to time series analysis: non-linear) within-variable
associations.

Claim 6
Network science is evolving into a mathematical science in its own right.

Throughout this section we have advanced the argument in mathematical and

quantitative or combinatorial terms, so as to make the point precisely, but we hasten
to add that the same considerations apply to qualitative network approaches. A
qualitative approach, perhaps even more so, is sensitive to the issues of connectivity,
systematicity, and dependence. For network science, the point is not whether one is

Downloaded from https://www.cambridge.org/core. Pontificia universidad catolica de valparaiso, on 22 Mar 2018 at 18:28:55, subject to the Cambridge Core
terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/nws.2013.2
 12 U. Brandes et al.

researching in a qualitative or quantitative way, but that the understanding of the

phenomenon treats relational connectivity and dependence as central.

4 The emerging science of networks

In light of the above discussion, we hope that this journal will provide a shared intel-
lectual space for network scientists working in many different fields to communicate
with each other about relational data.
To get there, we must recognize our various shared and disparate histories,
recognize that this field is quick-evolving, commit to compatible languages about
networks, and be willing to speak outside narrow disciplinary interests to broader
communities of scholars. The benefit, we believe, will be well worth the effort.
As editors of a journal attempting to encompass a broad field with a long and
storied history, we have already rejected the idea that network science “began” with
some kind of new discovery or even a Kuhnian paradigm shift tipped off by work
originating from physics, no matter how interesting or influential. Network science
is neither tied to nor “owned” by any other field.
We should not be ignorant of the forebears of our emerging science, and decades
of empirical research. The past 15 years have seen a boom of interest in networks
that does not overtly trace its roots to, for example, the sociometry of Moreno or
the sociology of Simmel. Even this older tradition has long borrowed from other
fields such as graph theory, physics, or statistics as it has developed.
Neither are these the sole progenitors of what we now recognize as network science.
Many streams of research are converging to create this new flow. The economic
problem of transshipment, or finding the most economical routes to transport goods,
is one of these streams. Seriation, or a method of relative dating of archaeological
artifacts is a 19th century splash that turned into a stream. Gene regulatory networks
have their own distinct research history since the late 1960s and have led to the
development of specific network models to account for transcription rates. And
there are many more examples showing that network science does not begin with
the advent of networks as models for complex systems, but as a perspective focusing
on interdependent relations that was developed in many areas.
The interest in models and tools applied to increasingly available large data sets
brought to bear by physics, biology, and socio-technical networks has brought with
it methodological and technological innovation. There is so much opportunity to
draw from work across fields, and so much history to be aware of. The journal
values work that points out the various streams of research that have informed
a particular field so that the intellectual relationships that shape the network of
network science can be made visible.

Claim 7
Network science is itself more of an evolving network than a paradigm expanding
from a big bang.

It is our intention to help the field of network science develop a canon of research
moving forward by publishing the most promising and widest-reaching work in the
field.

Downloaded from https://www.cambridge.org/core. Pontificia universidad catolica de valparaiso, on 22 Mar 2018 at 18:28:55, subject to the Cambridge Core
terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/nws.2013.2
 Editorial 13

This goal transcends disciplinary boundaries but we do have disciplinary goals

as well. Our major fields of editorial coverage (with area editors in parentheses)
include information science (Adamic), computer science and mathematics (Brandes),
communication, engineering and management (Contractor), economics (Goyal),
political science and psychology (Robins), public health and medicine (Valente),
physics (Vespignani), and statistics and sociology (Wasserman). Each editor has
identified key topics and debates within their area that they would like to see
addressed in the coming issues of Network Science and that list follows this editorial
as an attachment. Consider these an open call for work, but also consider Network
Science as welcoming of work that pushes this new science forward.
We are excited by the prospects of this new journal, Network Science. We
believe there is a distinctive science of networks that crosses traditional disciplinary
boundaries. It is ready to be brought together in a coherent form that transcends
disciplinary silos. We encourage all our readers to contribute to the journal to help
achieve these goals.

Acknowledgments
Part of the research that led to this editorial was funded by Deutsche Forschungs-
gemeinschaft under grant Br 2158/6-1 and the Social and Cognitive Networks
Academic Research Center of the US Army Research Laboratory.

Notes from the area editors

Lada Adamic, editor for information science
Information is an interdisciplinary field, just as network science. Therefore, a broad
range of topics can fall under this heading, including networked information (e.g., the
web, Wikipedia, citation networks), information dynamics in online, organizational,
and other social networks, and networks that can be constructed by representing
relationships between data (e.g., health, scientific, or historical data). We invite
contributions that include novel theoretical models, empirical studies, and methods
and applications pertaining to information networks.

Ulrik Brandes, editor for computer science and mathematics

We invite articles presenting original research in structural and computational
network science. This includes the study of network representations, algorithms,
data management, and visualization. A typical theory paper uses graph theory,
combinatorics, algorithmics, machine learning, information retrieval, or computer
graphics methods, whereas a systems paper concentrates on design aspects, im-
plementation, and performance assessment. Novel uses of network approaches in
application areas, and in particular those relating to social media, may also be
suitable for the information science area.

Downloaded from https://www.cambridge.org/core. Pontificia universidad catolica de valparaiso, on 22 Mar 2018 at 18:28:55, subject to the Cambridge Core
terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/nws.2013.2
 14 U. Brandes et al.

Noshir Contractor, editor for communication, engineering, and management

Management. The last few decades have witnessed the emergence of new organiza-
tional forms based on fluid, dynamic multi-level socio-technical networks. We invite
contributions that advance our understanding of the emergence and outcomes of
these novel forms of organizing both within traditional and informal intra- and
inter-organizational contexts. We also invite submissions that investigate how we
can use these insights to design more effective networks. In all of these areas we seek
to publish a broad range of articles that make significant theoretical, computational,
empirical, and/or methodological advances.

Communication. Digital technologies have significantly enhanced the modalities and

affordances we use to communicate at the interpersonal, group, organizational,
societal, intercultural, and global levels. These new developments prompt a reconsid-
eration of extant—and exploration of new—communication theories and methods.
We invite submissions that utilize novel network perspectives to advance our multi-
level understanding of the antecedents and outcomes of communication structures
and processes. In all of these areas we seek to publish a broad range of articles
that make significant theoretical, computational, empirical, and/or methodological
advances.

Engineering. Across various engineering disciplines there has been a growing interest
in understanding phenomena from a network perspective. These include contexts
as diverse as transportation networks, production networks, supply chain and
logistic networks, telecommunication networks, traffic networks, data networks,
mobile ad hoc networks, content distribution networks, peer-to-peer networks, sensor
networks, neural networks, nano networks, and regulatory networks. In all of these
diverse contexts, we seek theoretical, methodological, computational, and empirical
contributions that examine engineering issues such as network architecture, flows,
protocols, reliability, performance, optimization, routing, and congestions.

Sanjeev Goyal, editor for economics

In the field of economics, we would like to publish articles which explore the
economic origins and consequences of networks. We are also keen to publish articles
which explore network themes that lie at the intersection of economics and other
disciplines such as computer science, physics, statistics, psychology, and sociology.

Garry Robins, editor for psychology and political science

Psychology. We invite articles based in social psychology, social relations, and social
cognition but within an explicit social network or social system framing. We are also
interested in the structure of brain networks, including the network-based modeling
of empirical neuroscience data pertaining to brain connectivity; the structure of
cognitive networks (e.g., memory associations); organizational psychology, with an
emphasis on individual outcomes within a network-based organizational system,
or on the structure and outcomes of different types of organizational systems;
leadership within network-based social systems; small group studies of networks;

Downloaded from https://www.cambridge.org/core. Pontificia universidad catolica de valparaiso, on 22 Mar 2018 at 18:28:55, subject to the Cambridge Core
terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/nws.2013.2
 Editorial 15

and interaction of individual attitudes, traits and behaviors, and social network
ties, including network-based social influence. Finally, we are also interested in the
perception of social networks, network structures typical of different age groups,
or of other social categories; network-based social support and mental health; and
social networks and culture.

Political science. We hope to see theoretical, empirical, or computational studies

relating to network-based political science, policy networks, network governance,
including management of environmental systems, and dispersion of political atti-
tudes or policy-related behaviors across community social networks. Also welcome
are articles about health system networks, international networks linking nations,
historical network analysis, networks relevant to social movements, network-oriented
social media, and hyperlink analysis of political issues.

Thomas Valente, editor for public health and medicine

We invite manuscripts addressing all aspects of how networks relate to health, well-
being, and disease across all ecological levels and/or environmental systems. Original
scientific studies, as well as reviews are welcome. We also invite manuscripts that
develop novel measures, algorithms, perspectives, or insights that relate to health
and disease incidence, prevalence, progression, or transmission.

Alessandro Vespignani, editor for physics

Physicists specialize in the study of complex systems, both theoretical and applied.
The methods of physics are well suited to the problems arising in such systems; if
these systems happen to involve relational information, then network science benefits
from this research. We invite manuscripts from physicists studying networks, either
empirically or theoretically, particularly those interested in properties of networks
as a whole—shape, signature, or topology. Physics journals have published much
of the foundational work in network science over the past 15 years; we hope that
Network Science will be seen as a proper venue for such research.

Stanley Wasserman, editor for sociology and statistics

Sociology. We invite manuscripts using sociological theory and methods to study
network relationships among social actors, and the patterns and implications of
these relationships on groups and society. Network analysis is of course rooted in
sociology and social psychology—many advances over the past decades have come
from researchers in these fields. We hope that Network Science will be a leading
publication venue for sociological structural scientists.

Statistics. Network science is inherently a discipline rooted in data. Since statistics is

the science of data, we welcome research on both statistical theory of networks and
the statistical modeling of data arising from network studies. Network theories have
advanced over the decades because of development and quantification of hypotheses
that can be studied with statistics.

Downloaded from https://www.cambridge.org/core. Pontificia universidad catolica de valparaiso, on 22 Mar 2018 at 18:28:55, subject to the Cambridge Core
terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/nws.2013.2