|
Towards a Network Approach of Human Action
Theoretical concepts and empirical observations in media organizations
Abstract
This paper argues that network approaches can be helpful in describing
phenomena in the media. It presents data from an empirical observation
study in the newsrooms of German online media. We found surprising
similarities in the coded material from this observation. This leads us to
the conclusion that there are associations and sequences in human action
which can be analyzed on the basis of network theory. We therefore develop
a relational theory of human action.
Towards a network approach of human action: Theoretical concepts and
empirical observations
Towards a Network Approach of Human Action
Theoretical concepts and empirical observations in media organizations
Contact:
Thorsten Quandt, M.A.
Institute of Media and Communication Studies
Technical University Ilmenau, Germany
P.O. Box 10 05 65
D-98684 Ilmenau
Fon: +49/3677/69-4669
Fax: +49/3677/69-4695
E-mail: [log in to unmask]
AV needs: Power Point presentation, Windows PC/Laptop with CD-R drive and
projector
Paper submitted to the
Communication Theory and Methodology Division
Towards a Network Approach of Human Action
Theoretical concepts and empirical observations in media organizations
Short Abstract
This paper argues that network approaches can be helpful in describing
phenomena in the media. It presents data from an empirical observation
study in the newsrooms of German online media. We found surprising
similarities in the coded material from this observation. This leads us to
the conclusion that there are associations and sequences in human action
which can be analyzed on the basis of network theory. We therefore develop
a relational theory of human action.
Full Abstract
Although having a notable history in mathematics and the social sciences,
it was not until recently that network approaches have become increasingly
popular in the scientific community. Meanwhile, network approaches have
successfully been applied to such diverse fields as the modeling of
consumer behavior on the Internet, the search for DNA code or the
uncovering of terrorist groups. Nevertheless, network approaches are quite
uncommon in communication studies.
This paper argues that network concepts can be helpful in describing
phenomena in the media. It therefore sketches the framework of a relational
theory of human action and presents data from empirical observations in the
news rooms of media organizations, which have been carried out on the basis
of such an approach. During this 10-week project, we observed the behavior
of six German online journalists and coded more than 11.000 of their
actions. We found surprising similarities in the coded material which leads
us to the conclusion that there are a number of associations and sequences
in human action which can indeed be described and analyzed on the basis of
network theory.
1. Introduction: Networks and social theory
Network approaches are attracting a lot of attention these days, and in
particular from the general public. Just after September 11th 2001, the
idea of networks has been widely discussed, primarily in reference to
terrorist groups. Data mining algorithms based on networks algorithms have
been applied in the search for Al-Qaida members. Similar mathematical
models are used to identify consumer behavior on the Internet or patterns
in the DNA code. On a more general level, network metaphors have been used
to characterize modern society as a whole, even in newspaper articles and
on TV. While many of these discussions are based on popular network ideas
(and linked to similar phenomena like „the Internet"), some ideas actually
stem from an academic debate that took place in the recent years.
There, one can identify several sources for such a discussion. The two
major sources are:
(1) Mathematical concepts of networks derived from graph theory
(2) Sociological concepts based on the network metaphor
In the second case, the central term 'social connectivity' refers to a
broad understanding of society being similar to a network – which arguably
means: a network of interlinked agents (i.e. individuals or groups).
Especially in media and cultural studies, some researchers focus on the
role of media in connecting such agents.
While such an approach might be helpful in analyzing the relations between
people and the media, it is not the only conceivable way to apply a network
concept to human society (as explained in section 2). As an alternative way
of employing network concepts, we want to present the idea of a network of
action – a network that ultimately shapes the way we perceive and construct
the world (section 3). We will argue that, on the basis of our individual
actions, structures are emerging which can most likely be described in
terms of a network of meaning. This theoretical concept is supported by
data from an observational study of online journalism (section 4). There,
it became quite evident that human actions may be characterized by a
network of action elements, and also that suitable raw data taken during
such observations can be analyzed by means of standard network analysis
tools. In the last section, we will summarize the pros and cons of this new
way of theoretical and empirical thinking suggested here (section 5).
2. Network approaches: Some roots
Network approaches are not as new as the current debate would lead us to
believe. The concept of people forming a network is indeed an old one, and
it was first introduced to sociology by researchers like Georg Simmel and
Alfred R. Radcliffe Brown in the late 19th and early 20th century. They
used these ideas to describe social phenomena and structures, but mainly on
a metaphorical level. Empirical work, like the ethnographic studies of John
A. Barnes on kinship and social structures pushed the sociological concept
further ahead beyond its mere metaphorical meaning. Since then, the
theoretical concept of networks in sociology and social sciences has been
improved upon in many ways. In sociology as well as in economy, networks
became a central concept for the description of structured phenomena:
Williamson used this term to characterize a very efficient way of economic
coordination, Perrow discussed the distribution of power and influence
with the help of the network idea, Windeler applied the concept to
organizations, and just lately Castells presented his vision of a network
society, which has been discussed at lot since then, even outside the
scientific community. These were just a few examples (cf. , for a large
overview of standard texts on networks).
On the other hand, there is another major field of network approaches which
can be derived from the so-called "graph theory". The latter is the logical
and mathematical basis for the formal description and analysis of networks
and connections. A graph is a general type of structure which can be
represented by elements (nodes) and its connections (links). The beginnings
of graph theory date back to the late 18th century, starting with Leonhard
Euler and his solution of the so called "Königsberg problem" . The
mathematical graph theory was later refined in terms of a complex network
theory, borrowing some ideas from chaos theory and the analysis of
self-organizing systems . With the increasing power of computer software,
this kind of network analysis is becoming increasingly popular in many
areas of research, ranging from the decoding of the human genome to the
analysis of organizations or the uncovering of terrorist groups. The
standard numerical tools include data mining packages and the application
of artificial intelligence based analysis algorithms (cf. .
It is not surprising that the applications of graph theory are manifold,
due to the logical (and therefore empirically empty) quality of such a
point of view. But it is quite of a surprise that the sociological point of
view concerning networks is somewhat conservative when it comes to choose
the types of phenomena that it should describe. Or to say it more clearly:
the choice of network nodes. In most cases, the sociological network
approach refers to society or groups as structures being similar to a web –
forming a network of interlinked agents (individuals or groups). So the
elements or nodes that appear in sociological network theories are human
beings.
We would like to argue that this approach is far too narrow, and that
network concepts can be applied to other social phenomena as well,
especially to the basic category of human action. Actually, various
companies on the Internet are already operating in the same direction. They
try to model buying behavior using network algorithms: the nodes are the
individual buying acts, which are connected to other buying acts, and in
the end, there emerges a complex network of connected buying acts. This
structure is what these companies are actually looking for in order to be
able to predict consumer behavior. And there is already a general term for
this kind of analysis: It is called "data mining" or "knowledge discovery"
(cf. ).
In this paper we will argue in the same direction: we will not focus on a
network of interlinked agents, neither on individuals or groups, but on
networks of actions instead. Which does not mean that we want to leave out
human beings, or that we want to suggest that networks of individuals or
groups would not be a helpful concept. But we believe that choosing them as
network nodes might not be the only promising way of applying network
theory to social or cultural phenomena. In the following section, we would
like to present some theory that will support our point of view.
3. Theoretical background: Towards a network approach of human action
When describing human actions, it cannot be enough to just label the types
of individual acts that are being performed by a certain person (asking
„what is s/he doing?"). There are a number of factors that determine the
way in which these acts are finally embedded into the flow of action. For
example there is the time and space framework („where and when does s/he do
this?"), contact persons or relations among subjects („...in contact with
which person?"), the material resources („...with the help of what type of
resource?") and the general sense making ‚location' of the act („...in
which context?"). These elements may be looked upon as constitutional for
human actions, and most of them have already been identified in the
standard works on a sociological description of human action (cf. , and in
more recent publications like Giddens' structuration theory . While we can
surely conceive of other elements as well, the elements described here may
already be sufficient to characterize individual actions.
A figure illustrating these interconnected elements is shown below (fig. 1).
[--- ??? Graphic Goes Here ---]
Fig. 1: An individual act as a star network of associated elements
In this figure, we clearly observe a network structure, but for good
reason: The constituent elements of each act are linked by the action
itself, and therefore they constitute a small star network. Without its
central node, the network would cease to exist, whereas some of its outer
elements might eventually be missing under certain circumstances (which is
true for relations among subjects and resources that are not essential for
each and every individual human action).
On the other hand, human actions do not exist as moments frozen in time.
Instead, they are part of a constant flow in time, with one act followed by
another. In our everyday life, we constantly do something, followed by
another action, and based on a certain history of acting. Such a history of
action is only possible because we perceive actions as being related to
each other, and in particular when they take place in sequence. For
example, we may assume that the majority of journalists know which steps to
follow when they have to write an article: They know their sources for
research, they remember possible starting points from earlier work on
similar topics, they know when they have to talk to somebody, and they know
when they should stop researching and begin with writing things and so on.
They obviously remember single micro-steps as well as large coherent
sequences. In common language we call that "experience". Sociologists and
psychologist alike assume that humans remember actions through cognitive
processes by means of what Schütz called a "stock of knowledge at hand" .
Which is a repertoire of basic rules being at our disposal in order to
develop strategies for our future actions. In the language of the network
theory, these rules are operating as connection rules, because they are
able to describe the structures among various action elements like
resources, types of actions, personal contacts, contextual information and
the space and time framework itself. Therefore, this stock of knowledge is
basically a huge network of relations which constitutes human memory and
which lays the foundations for further human actions, thereby creating the
very identity of a person performing those acts.
[--- ??? Graphic Goes Here ---]
Fig. 2: Sequences of actions are being transferred to the stock of knowledge
Now what happens if several individuals, for example journalists in
(different) newsrooms, have contact to similar subjects, similar resources,
working under certain material conditions, and being confronted with
similar actions? First of all, they will build up similar relations among
certain action elements in their stock of knowledge. That does not
necessarily mean that they are forming similar traces of memory in their
brains. Actually this is highly unlikely, because the perceived actions
usually relate to different elements in each individual stock of knowledge.
But the important thing is that these subjects share the same relations.
Let us take this paper as a simple example: As a reader you will perceive
our words in one way or another. And the way in which you relate the
information contained in this paper to your actual knowledge is a highly
individual process, because we are all entering such a process with rather
different memory structures. Nevertheless, you will share your relation to
this paper with any other reader, even if s/he is thousands of miles away
and lives in a totally different living environment.
So while the nature of links might vary, their relational qualities will
basically be the same. They will also stay the same if people share parts
of their stocks of knowledge through communication or through
co-orientation.[1] There may not be direct contacts between all the initial
action elements, but at least there remain some links. For example if you
(as a reader) would tell a friend what is explained in this paper, say in a
few days, this friend would most likely share a – somewhat weaker – tie to
the present paper (cf. fig 3).
[--- ??? Graphic Goes Here ---]
Fig. 3: Building networks of meaning through shared relations
So through their everyday's practice, and through similar connections among
actions, resources, contexts etc., people will build up comparable webs of
sense-making relations (i.e. connections that allow for a complex behavior
which creates options for subsequent human actions) – therefore, the
individuals are actually sharing some 'meaning' (at least to a certain
amount).
4. Empirical applications: Observational study of online journalists
4.1. Design of the study/Methodology
One of the advantages of network approaches is that they can easily be
applied to empirical studies: After defining appropriate nodes and the
relations among them, the structure of these networks may simply be
described by graphs (which means that they form a logical structure which
can be translated into a formal/mathematical language; cf. , and . The
above-mentioned approach already provides us with the basic elements that
may serve as categories for such empirical studies: The 'nodes' of
individual acts can be operationalized for direct use in (observational)
studies. Surely, such an approach may also serve as the basis for surveys,
but observations seem to be most natural way to analyze human action/behavior.
Based on the theoretical approaches mentioned above, a large observational
study could actually be realized. During a 10-week study in the newsrooms
of 5 German online newspapers, the actions of 6 online journalists have
been observed. The motivation for such a study was the idea that there
might be some sort of professional rules evolving for this new area of
journalism. At the time of the study, German researchers did do not know
very much about 'real life' working conditions and the rules of online
journalism . Therefore a closer look at the structures (the elements of
actions and relations among them) shaping the everyday work of a journalist
was certainly overdue.
The operationalization of the individual action elements (types of actions,
context, space, time, resources, subject relations) resulted in a codebook
containing about 250 numerical and symbolic codes that had to be memorized
by the observer. During our observations, the flow of action was broken
down into individual acts[2] and the acts themselves into the constituent
elements which were itemized in the codebook (the graphics shown below
should give us an impression of how this was done in principle, cf. fig. 4;
different conditions/values of the individual elements are indicated by
different shapes).
To provide a better understanding of the working environment, observational
diaries were set up to write down open questions which could be answered
during eleven interviews with the journalists and their editors in chief.
In addition to that photographs of these workplace were taken and ground
plans of the work places were drawn in order to get some impression of the
working conditions of the journalists. But the core results were the coded
observations. We obtained a data matrix with 11.671 acts (corresponding to
483 hours and 28 minutes of observation); each act consisted of about 50
variables that would describe its constituent elements in more detail.[3]
Therefore the data basis for further analysis was huge.
[--- ??? Graphic Goes Here ---]
Fig. 4: Breaking down the flow of action into a data matrix
The aim of the study was to identify patterns among the relational data
contained in this matrix. Or to put it in other terms: the aim was to find
similarities and rules of action that might be typical for online
journalists. Such patterns (work routines, rules of action) can evolve into
different directions – first, there are frequent connections between
different elements, which are called associations. Or one might find
temporal patterns, which are called sequences. The main question concerning
the latter type of connection was the following: Are there certain actions
that follow other actions on a regular basis?
4.2 Results
4.2.1 Overall distribution of journalistic action
The results of this study reveal some striking similarities in the observed
actions of the different journalists, although there were no direct
contacts between the observed individuals – they all worked for different
media organizations in different towns. Still the schedule of their working
days, as well as their general rules of action and their use of resources
(including technological devices) followed comparable patterns. There
appear to be invisible ties between the individuals and their actions –
which is "net-work" in both sense of such an expression. This is quite
astonishing, given the fact that online journalism is a relatively new
field, with no original tradition of its own that would make it different
from journalism as a whole. Even the literature on this subject has not
identified special rules of online journalism (although there really is a
bulk of articles dealing with online journalism). The above mentioned idea
of human actions as being shaped through the network of sense-making
relations seems to be useful in explaining this fact: It is assumed that
the similarity of relations leads to the formation of comparable structure
building processes in the stock of knowledge, as well as among the observed
actions.
Some empirical data from our study will give us an impression of the
above-mentioned similarities. First of all, the overall distribution of
types of actions was similar for almost all of the observed journalists
(with the exception of one journalist who had a lot of technical tasks;
this was actually due to the fact that he was the only online journalist in
his media organization).
[--- Pict Graphic Goes Here ---]
Fig. 5: Overall distribution of observed actions[4]
The above pie graph (cf. fig. 5) shows the overall distribution of time
spent on different actions during a journalist's office hours. The biggest
pieces are research, text production, interpersonal communication, the
communication through media, and production jobs. That is roughly what one
would expect from a journalist, although the high level of communication
looks rather surprising (which mostly consists of co-ordination with
colleagues, though – for example through organizational talks).
It is interesting to see the homogeneous distribution pies related to
different journalists. The following viewgraph (cf. fig 6) compares the
amount of time spent on the individual actions for two journalists of the
Netzeitung in Berlin (NZ1 and 2), the Frankfurter Allgemeine Zeitung (FAZ)
and the 'Tagesschau' in Hamburg (TS).
[--- ??? Graphic Goes Here ---]
Fig. 6: Comparison of distribution pies (observed actions) [5]
There are obvious similarities between the journalists from FAZ and
Netzeitung. The distribution pies for the editors of the Netzeitung
actually look as if they were Siamese twins. So both of them are doing
almost the same things and spend approximately the same amount of time on
similar actions, although they clearly are two different individuals.
[--- ??? Graphic Goes Here ---]
Fig. 7: Workplaces of online journalists
Another example: The above photographs (cf. fig. 7) depict the workplace of
two online journalists. They look very similar: A lot of printouts, and two
flat screens. The journalists used the two screens for just the same
reasons (content management system on one screen, agency news on the
other). It is worth noting that in both cases, the flat screens were bought
by the media companies because the journalists asked their management to do
so. So that was not a structure that shaped the journalists action in the
first place, but they would ask for this setup due to their working
necessities. The most surprising fact however is that the pictures were
taken at two rather different media, namely the FAZ in Frankfurt, and the
Tagesschau in Hamburg. The main company of Tagesschau online is a public TV
station like the BBC, while the main company of the FAZ is a conservative
nationwide newspaper.
So we conclude that some individuals in the observed newsrooms have
obviously developed comparable working patterns, and that they are using
comparable resources in similar working places.
4.2.2 Associations and sequences
According to the above-mentioned theory, some actions refer to certain
resources, resources to places or time frames, time frames to actions and
so on. These relations can be described by associations and sequences
contained in the data matrix. With the help of the standard data mining
program Clementine, we carried out a network analysis of these
associations. In principle, such an analysis counts the connections between
the individual values of the coded variables and compares the actual number
of observed connections between two values with the overall number of
connections of the first value. Therefore it gives us an overall impression
of the strongest connections (for example, it will give you an impression
about the strength of the ties between certain actions and resources). The
network viewgraph shown below contains all the action types (on the left)
and resources (on the right) that were observed during our study (cf. fig. 8).
[--- ??? Graphic Goes Here ---]
Fig. 8: Association analysis (action type x resources)
Obviously, these connections are not evenly distributed. There are some
strong ties, and a lot of weak ties – and quite a lot of nodes are not
connected at all. A change of threshold within Clementine's network
analysis algorithm will highlight the most frequent connections and delete
all of the weaker ties (cf. fig 9).
[--- ??? Graphic Goes Here ---]
Fig. 9: Association analysis, strongest connections (action type x resources)
As the number of connections is shrinking, one is finally left with just
the strongest ties – which are of course very obvious links, after all. For
example, the communication acts are very strongly related to the resource
'telephone'. That hardly comes as a surprise. Nevertheless, it is also a
clear indication that the resources were well defined, that that they just
serve one major purpose. Content management systems, on the other hand, are
of a rather different nature. They are used as central nodes for many types
of actions. This may indicate the importance of such tools for the
production of news in online journalism. It is also obvious from the
network diagram that action types always refer to the same arrangement of
resources (one action leads to one resource which leads to another action
etc.). These relations create robust sense-making patterns, because
resources and actions are really glued together by such links.
Another type of analysis focuses on the temporal sequences of individual
acts over time. While this can be carried out with the help of sequence
analysis algorithms as well (like Clementine's CAPRI algorithm), we chose
to carry out a graphical analysis first. As Keim notes, graphical analysis
by a human being can be superior to computer algorithms, simply because
humans easily detect certain patterns on the basis of their huge knowledge
of similar observed phenomena. The granulation of the observation is a
difficult problem for computer programs ("what is the size of the elements
that should be observed, how long should the sequences be etc.?"). Similar
problems appear when it comes to the interpretation of raw data ("what kind
of sequence is trivial, what kind of sequence is important?").
In order to analyze those sequences, we applied a "slicing" algorithm to
the material, cutting the observed actions into 5-second pieces (the
starting point and the end point of each action have been coded). The
resulting data consisted of temporal cases, where each case represents an
equal amount of time. Based on this transformed data set, it was possible
to produce a graphical display of actions over time that may form the basis
for further analysis (cf. fig 10 for an example).[6]
Fig. 10: Time based graphical analysis, "piano roll" graph (for action types)
The picture represents one working day. Each row shows one category of
actions. The vertical lines depict the starting/end points of different
phases of a working day: In the beginning (8.30 - 9.50), the online
journalist is reading a lot of e-Mails (media based communication) and does
not write very much. Which is something like an orientational phase that
marks the beginning of almost every working day. The following bigger "work
phase" shows quite a lot of research in the beginning (which also serves a
first orientation to collect interesting news), with more writing during
the second half of this period. Then there is a break (13.30 - 14.30).
After the break, there is a second working phase with almost no movements,
but phone calls, e-mail exchanges and long sequences of research. The last
period of the day is characterized by writing and researching articles
(those actions are usually bound to one news article/topic). There is
almost no phoning/media based communication going on (only a few contact
persons were available after 16:00, although there was a big interview
taking place afterwards, in this special case), and this period is
dominated by long sequences of writing .
This is quite a common pattern for online journalism. Obviously there is a
constant stream of writing and research happening during the working day.
There are no real production deadlines, but a constant need for researching
and reworking news. Nevertheless, some of the communication processes seem
to fade out by the end of a working day, which is dominated by writing. And
despite the fact that there are no real production deadlines, we still
observe orientational phases at the beginning of each working day and
production peaks during the day. External contacts also seem to shape
working patterns to a certain extent.
Without going into the details, it may be noted that this finding is in
clear contrast to various speculations which claim that online journalism
may not be bound to the restriction of time.
A detailed analysis of associations and sequences among our data (remember
that we can go down to details of five seconds) also shows that there are
interesting work patterns which seem to develop into rules of action. For
example, writing and research routines seem to follow similar patterns in
the vast majority of all cases, with a consistent use of content management
systems and satellite/internet based news agency information as basic
working resources.
This may illustrate some of the possibilities of the theoretical and
empirical approach. In a last section, we will summarize the pros and cons
of the network perspective, and draw some final conclusions.
5. Conclusion: Theoretical and empirical potential
When using a new approach, one has to ask: What might be the advantages,
what might be the disadvantages of such an approach, compared to
traditional ways of looking at society and action? In that sense, there are
certainly some disadvantages to the network perspective:
• The theoretical approach is only loosely tied to traditional perspectives
about journalism.
• It offers no simple, singular description of phenomena in the media.
• It depicts complicated, fluctuating relations that may lead to
ambiguities and sometimes even to contradictions.
Nevertheless, there are also advantages to such a procedure:
• It takes into account the complexity of the social construction of reality.
• As an analytical approach to the production of action networks and
meaning in every day action (similar relations of elements), it gives us
more than just a handful of metaphors for describing social phenomena.
• It is an inherently dynamic view, which is helpful if you want to look at
changing aspects of social life.
• It is open for empirical research.
Empirically, the network-based observation of human action shows its
potential when it comes to a detailed description of working behavior. It
can certainly be carried out in addition to surveys, as a supplement and to
correct certain aspects. One has to note however that observational studies
cannot be carried out on a representative basis, because they are just too
costly and would interfere with the editorial processes if carried out on a
large scale. But they can be employed to create empirical conjectures,
uncovering unknown relations with the help of data mining tools.
These tools have many applications beyond the few examples shown in this
paper. Algorithms based on (neural) networks could lead to a deeper
understanding of human behavior. We believe that for journalism research,
this could open up a new way of theoretical thinking as well as new ways of
empirical research and analysis.
References
(Translations in parentheses)
[1] The similarity between those relations can be explained through
co-orientation, i.e. orientation towards similar phenomena, and also
through communication. However, similar structures are neither a necessary
nor and an exclusive effect of communication. The latter can be described
as a special type of action that transfers parts of 'ego's' memory
structures into the stock of knowledge of 'the other'.
[2] Acts were defined as being interconnected and coherent. Thus an act
would end when at least one of its elements had changed. The question about
the observed size of acts (the „granulation" of observation) is not
answered by such a procedure – but this is was not the central question
when we were looking for patterns, because relationships will be visible
even when the size of observed acts does vary. The relational structure
will actually stay the same .
[3] The high number of variables per case is partly due to individual
coding for up to four resources, four (groups of) contact persons and
several context variables.
[4] Numbers for the smaller pieces of the pie have been omitted for the
sake of clarity.
[5] All numbers have been omitted for the sake of clarity.
[6] One of the biggest problems was the possibility of multiple actions
taking place at the same point of time. But the „slicing" of the data would
allow for the transformation of an action-based matrix (1 case = 1 action)
into a time-based matrix (1 case = 1 time step, with new variables
describing all the actions at this point of time).
|
