close

Вход

Забыли?

вход по аккаунту

?

j.socnet.2017.10.002

код для вставкиСкачать
G Model
ARTICLE IN PRESS
SON-1053; No. of Pages 11
Social Networks xxx (2017) xxx–xxx
Contents lists available at ScienceDirect
Social Networks
journal homepage: www.elsevier.com/locate/socnet
Heating up the debate? Measuring fragmentation and polarisation in
a German climate change hyperlink network
Thomas Häussler
Institute of Communication and Media Studies, University of Bern, Fabrikstrasse 8, 3012 Bern, Switzerland
a r t i c l e
i n f o
Article history:
Available online xxx
Keywords:
Brokerage
Exponential random graph model
Polarisation
Climate change
Hyperlink
Contentiousness
a b s t r a c t
Research into polarisation on the internet has so far primarily focused on contentious issues and yielded
contradictory results. Shifting the focus to a non-contentious setting, this article combines community
detection with brokerage analysis and exponential random graph models to assess the degree of polarisation at different levels of a German hyperlink network on climate change. Although homophily accounts
for a moderate degree of polarisation at the top level of the network, the communities reveal that other
factors prove more decisive in shaping its structure and the article thus contributes to a more refined
understanding of the nature of online polarisation.
© 2017 Elsevier B.V. All rights reserved.
1. Introduction
One of the central questions of current research regarding the
internet is whether it is able to expand the political public sphere.1
Optimists argue that the low costs and easy access are apt to generate a more level playing field by granting a greater voice to civil
society actors, which in turn should make political discourse more
diverse as it integrates a broader range of (often marginalised)
views (e.g., Bimber, 2003). Pessimists in turn point out that the
radical democratic potential of the web is undermined by virtue
of its normalisation—that is, the fact that on closer examination it
is equally stratified, demonstrates the same distortions of power,
and therefore reflects, rather than complements, the reality of the
offline world (e.g., Hindman, 2008).
What remains uncontested is that the internet and its hyperlink
topology add a new dimension to the way actors become visible,
how they relate to one another through their social and political
‘acts of association’ (Rogers, 2012; p. 6) and thereby structure the
public space, and how we conceive of it. Indeed, the algorithms of
search engines such as Google take the hyperlinks running between
websites as a measure of their visibility and prominence. Just what
the effects of these individual choices are on a larger, cumulative
level and how they affect the topology of the internet is a matter of
some debate and concern.
1
One of the main worries is that actors primarily link to
like-minded others, abetting the formation of ‘echo chambers’
(Sunstein, 2009), which effectively undermine any hopes of a
greater diversity associated with the digital revolution. This form
of polarisation is generally taken to consist of a sorting process,
through which actors with opposing viewpoints come to be located
in politically uniform camps that have little communication with
one another. The concern is that this separation between opponents is apt to erode the very structures of a digitally expanded
public sphere in the long run (Chaffee and Metzger, 2001).
Against these far-reaching implications, the existing research
on polarisation of the online space has for the most part examined linking patterns on single platforms such as Twitter (Williams
et al., 2015), the activities of single civil society actor types such
as bloggers (Adamic and Glance, 2005; Elgesem et al., 2015), and
generally focused on instances of heightened controversy between
the opposing camps of an issue. For all the important insights these
studies have generated, we are still left with an incomplete understanding of the dynamics at work: we know little about polarisation
processes outside individual online platforms, specific actor types,
or how the actors structure their hyperlink communication in less
contested issues. Furthermore, by focussing primarily on the general level of the network, the existing research tends to overlook the
more complex patterns that shape the intermediate level of brokered relationships and the local level of tie formation processes.
It also faces difficulties in distinguishing between the expected
degree of interactions between like-minded actors and instances of
E-mail address: thomas.haeussler@ikmb.unibe.ch
This article uses the terms ‘internet’, ‘web’, and ‘online’ interchangeably.
https://doi.org/10.1016/j.socnet.2017.10.002
0378-8733/© 2017 Elsevier B.V. All rights reserved.
Please cite this article in press as: Häussler, T., Heating up the debate? Measuring fragmentation and polarisation in a German climate
change hyperlink network. Soc. Netw. (2017), https://doi.org/10.1016/j.socnet.2017.10.002
G Model
SON-1053; No. of Pages 11
ARTICLE IN PRESS
T. Häussler / Social Networks xxx (2017) xxx–xxx
2
proper polarisation, which significantly distort the communicative
space.
This article therefore explores the degree to which linking patterns between different types of online actors reveal tendencies to
fragment the online space into polarised pockets by analysing a
hyperlink network on climate change that originates in Germany.
This also allows us to investigate whether polarisation is an irredeemable quality of any res publica or whether it relates to the
degree of contentiousness that surrounds an issue. Much research
tends to concentrate on instances of heightened political controversy, which may lead to a somewhat incomplete understanding
of polarisation processes, as we tend to ignore other configurations
of political issues. From a methodological point of view, current
research suffers from the general shortcoming that while several
studies have documented tendencies of online polarisation, it is not
clear to what extent this is really evidence for further-reaching segregation processes, or whether it simply reflects a normal economy
of attention between opposing camps (Benkler, 2006). This raises
the question of measuring polarisation and defining corresponding thresholds. Here, the present study argues that polarisation is
a complex dynamic that occurs at different levels of the network
and proposes a set of measures that capture polarising tendencies
on the general, intermediate, and local levels, assesses whether
they are beyond the threshold of statistical significance, and examines how substantive the effect is. Whereas the different network
levels offer a more detailed concept of polarisation processes, the
thresholds allow us to distinguish between regular communication patterns and segregationist tendencies and assess their relative
strength.
The article proceeds as follows: the next section reviews the
literature on fragmentation and polarisation on the internet, highlighting the conceptual issues with which it has been hampered.
This is partly due to a one-sided focus on highly contested debates,
and we will therefore widen the theoretical focus to encompass
both ‘open’ and ‘closed’ issues. The methods section presents the
case selection, outlines the procedure used to generate the German
climate change hyperlink network, and introduces the different
measures that allow us to assess the form and degree of polarisation, moving from the general level of the network and the
intermediate level of the relationships between the single communities to the local level of the nodes’ tie formation processes. As
the analysis shows, the overall moderate degree of polarisation is
mainly present in the higher levels of the network, and extending
the analysis to the local level by employing exponential random
graph models (ERGMs) safeguards us against premature conclusions. The concluding section embeds the analysis in the wider
context of the current debate and shows that closed issues, as do
open ones, pose a number of challenges for normative theory.
2. Literature review
Empirical research into polarisation in hyperlink networks takes
a different approach to the prevalent view, which is employed
above all in audience studies (e.g., Stroud, 2008) and is rooted in
the social psychological literature of the 1970s (for an overview,
see, e.g., Myers and Lamm, 1976). This recent work marks a return
to the ‘risky shift’ tradition (Stoner, 1961), or more generally, the
‘group polarisation phenomenon’, which has been documented in
numerous experiments and shows that subjects in groups tend
to adopt more extreme positions with regard to their individual
pre-group responses (Myers and Lamm, 1976). Polarisation in this
context refers to a shift of the views and attitudes of individuals
within a group towards one extreme, for instance, in terms of risk
assessment or political attitudes.
In contrast to this, communication researchers analysing hyperlink networks have focused on online communities formed by the
interaction between websites or blogs (e.g., Adamic and Glance,
2005) and are interested in whether the resulting topology leads to
enclaves of like-minded actors. Polarisation here refers to a process
that affects the structures of a networked public sphere (Benkler,
2006), as opposing views become segregated from each other and
thus undermine the porousness of the informal public (Habermas,
1996), leading to worries about its potential to integrate societies
and confer legitimacy to political decisions.
In its standard form, this type of polarisation is modelled as a
bidirectional process, which differs from the one-directional shift
of social psychological studies. Despite the differences between the
two approaches, it is easy to see that they complement each other,
as enclosed and ideologically homogenous spaces on the internet
are apt to act as catalysts to the individual polarisation of online
users, and the importance of authors such as Sunstein (2009) lies
in the fact that they have indeed connected the two perspectives.2
An important difference between the two notions regards the
analysis of polarisation. Whereas social psychologists have examined the effect of group discussions on the views and attitudes of
individual subjects, internet researchers have so far failed to study
hyperlink interaction as it unfolds. Rather than studying the process itself, whereby actors come to build more intense relationships
with those sharing the same position, they have focused on its product, that is, clearly delimited areas shaped by the interaction of
like-minded actors, which is taken to indicate a preceding sorting
process (e.g., Elgesem et al., 2015; Guerra et al., 2013; Hargittai
et al., 2008; Williams et al., 2015). In effect, what this research documents are the hyperlink fragmentation tendencies of the online
spaces that lead to politically homogenous enclaves that are segregated from one another. These are taken to reflect instances of
hyperlink polarisation, and the present study employs this definition when examining the structure of the German climate change
online space.
Several blog-linking studies empirically support the echo
chamber view. According to Adamic and Glance (2005), the US blogosphere fragmented before the 2004 presidential election into
liberal and conservative camps, a finding confirmed in part by
Hargittai et al. (2008) analysis of the linking patterns in liberal and
conservative blogs, although they also find substantive argumentative exchange between the opposing camps. Similarly, Woo-young
and Park’s (2012) study of the Korean blogosphere during the US
beef import controversy reveals a fragmented and polarised online
space. Conversely, Adamic (2008) shows that, depending on the
issue, here the bankruptcy bill, a usually divided blogosphere might
temporarily integrate across the political divide.
From the perspective of climate change as a political issue, two
recent studies partly confirm an existing polarisation: analysing 1.3
million blog posts in the English speaking blogosphere, Elgesem
et al. (2015) employ an inductive community detection procedure in combination with a topic-modelling approach to pinpoint
the distribution of topics in the discourse. They find a cohesive
community of climate-sceptical bloggers alongside several climate
advocate clusters. Contrary to this division, however, the topics
identified by the study cut across the different communities and
thus integrate the network. Applying a similar approach, Williams
et al. (2015) examine the climate change debate on the microblogging platform Twitter, restricting the analysis to English tweets.
The communities they identify through their linking patterns only
partly form uniform clusters characterised by a single view on
2
Note that the original social psychological literature does not require groups
to be homogenous with regard to the property of interest to the researcher—the
polarisation effect occurs nevertheless (e.g., Myers and Lamm, 1976)
Please cite this article in press as: Häussler, T., Heating up the debate? Measuring fragmentation and polarisation in a German climate
change hyperlink network. Soc. Netw. (2017), https://doi.org/10.1016/j.socnet.2017.10.002
G Model
SON-1053; No. of Pages 11
ARTICLE IN PRESS
T. Häussler / Social Networks xxx (2017) xxx–xxx
the issue. Significantly, their analysis also produces open fora,
where sceptics and activists interact, although the exchanges are
often characterised by negative sentiments about the other side.
Nevertheless, fragmentation and polarisation dynamics appear to
produce more complex results in practice than predicted by theory.
Despite the findings these studies have generated, research in
the area is inherently plagued by the difficulty of specifying proper
polarisation measures and thresholds. Given that homophily is
such a pervasive factor in the organisation of social and political life (McPherson et al., 2001), actors sharing the same position
on an issue are bound to interact more strongly with one another
than with those they oppose; exactly when such interaction patterns become detrimental to political discourse is difficult to gauge
and a matter of some dispute. Benkler (2006) correctly points out
that preferential interactions with like-minded others in many
instances follows a normal economy of attention and is indicative of internal discussion processes rather than of segregationist
dynamics.
Given the diverging results reported by the existing analyses, the
question of whether online communication furthers polarisation or
integration is far from settled. The present article ties in with these
studies but also attempts to expand them in important ways to
arrive at a more detailed understanding of the underlying dynamics. Methodologically, it develops a more fine-grained approach to
assess polarisation on three distinct network levels: first and in
line with Elgesem et al. (2015) and Williams et al. (2015), on the
general level of the network the analysis determines inductively
the extent to which the actors tend to build communities of likeminded actors; second, on the intermediate level it measures how
introverted or extroverted the communities are in relation to each
other, and to what degree these relationships significantly depend
on only a few brokering actors; and finally, it evaluates at the local
level of the single nodes whether their tie formation processes are
governed by an orientation towards like-minded others or whether
other factors play a more dominant role. This allows us to distinguish between different forms and intensities of polarisation.
In this view, ‘normal economies of attention’, while displaying
some degree of polarisation, remain below the threshold of statistical significance (see the measures of polarisation below) and
should lead to largely mixed communities that intensely interact
with each other through a variety of actors, who in turn form their
ties to others regardless of their political positions. The echo chambers of which Sunstein (2009) has cautioned, by contrast, differ in
these dimensions and are defined by sorting processes that lead
to statistically significant results in terms of uniform and antagonistic communities, which are largely closed off from each other,
where only a few actors mediate the relationships between them
and where nodal tie formation processes are substantially driven
by the orientation of the actors towards like-minded others.
At the same time and in line with Benkler et al., (2015), the
approach taken extends the focus from single actor types such
as bloggers or platforms such as Twitter to hyperlink issue networks that connect different actor types. Although focusing on an
issue network originating in one specific country, the approach
does not restrict the actors’ linking patterns to the national context
but explicitly takes into account the boundlessness of the internet
and the resulting transnational character of the online issue space
(see Elgesem et al., 2015). Finally, the article moves the attention
from institutionalised events such as general elections or phases of
heightened protest and mobilisation to the courant normal of politics. Shifting the perspective in this way allows us to assess the
degree to which polarisation is a constitutive element of political
debates. The contentiousness of an issue thus becomes a central
component of the study design, and the next section develops a theoretical account of it, from which we can derive a set of hypotheses.
3
3. Polarisation and the controversy of issues
Most of the studies reviewed concentrate on those stages in
an issue cycle that are marked by increased contentiousness, and
for good reason: it is here that we can examine in greater detail
how polarisation shapes the relationships between the actors in
the online space. As a result of this orientation, however, we know
little about the other issue configurations and consequently cannot tell how, and to what extent, controversy and polarisation are
related. By making the degree of contentiousness a central element of the analysis, we emphasise the context in which an issue
is embedded. While there are numerous dimensions that can be
taken into account, we focus on three of the most defining ones:
the resonance of opposing positions in public opinion (Converse,
1987) and in the coverage of the media (Zaller, 1992) as well as
the alliance structures that configure the relationships among the
actors (Kriesi, 2004a, 2004b). For the sake of analytic simplicity,
we distinguish between contentious and non-contentious issues
(Baumgartner and Jones, 1991).
We can define an issue as contentious when a political dispute is
‘open’ in the sense that there is an ongoing debate between (at least)
two camps who are visible in terms of their resonance in the media,
whose positions are reflected in the popular concerns of people and
public opinion, and who are embedded in an alliance structure,
which includes civil society organisations as well as members of
the political elite and/or the economy. This configuration describes
a political public sphere that is in flux, defined by dissent that cuts
across actor categories and social strata and divides the public. By
contrast, an issue is ‘closed’ when there is a far-reaching agreement
over public policy among the political (and economic) elite, when
the media affords space overwhelmingly to only one side of the
issue, when public opinion is one-sided, and when dissenters are
isolated. Note that this constellation differs from a non-issue, where
a debate has either been settled or the issue awareness is so dispersed that it fails to become visible on the public agenda. Research
has so far exclusively examined open issues, whereas we formulate
hypotheses below that capture some of the central aspects of more
closed debates.
3.1. Hypotheses
Based on the findings of the studies reported above, it generally
seems plausible to assume that hyperlink networks in closed issues
such as the one examined below should display lower degrees of
fragmentation and polarisation. The following hypotheses capture
the main dynamics of polarisation and proceed from the general
level of the network to the relationships between its communities,
and finally to the more specific aspects of node formation at the
local level.
The first hypothesis examines how the degree of contention
in an issue affects the cohesion of online hyperlink networks as
a whole. As the existing research shows, networks in open issues
tend to fragment along the lines of political conflict, although some
studies also find clusters in largely polarised networks that combine actors with opposite positions (Adamic and Glance, 2005;
Elgesem et al., 2015; Hargittai et al., 2008; Williams et al., 2015). In
closed issues, the degree of contentiousness is lower, one side of the
debate holds a hegemonic position, and to gain visibility those in
the minority are forced to connect to the mainstream of the debate.
Rather than producing echo chambers, these issues should therefore lead to open forums in the network, and the fragmentation that
might still occur should not run along the lines of the position of the
actors. Communities in the networks of closed issues should therefore display a similar composition in terms of the actor positions
(Hypothesis 1).
Please cite this article in press as: Häussler, T., Heating up the debate? Measuring fragmentation and polarisation in a German climate
change hyperlink network. Soc. Netw. (2017), https://doi.org/10.1016/j.socnet.2017.10.002
G Model
SON-1053; No. of Pages 11
ARTICLE IN PRESS
T. Häussler / Social Networks xxx (2017) xxx–xxx
4
Polarisation is often assessed in the literature based on the level
of the relationships between online communities of like-minded
actors (Adamic and Glance, 2005; Hargittai et al., 2008). Typically,
if the connections established within the communities outweigh
those that cut across the political divide, this is taken as an indication of polarisation. As a consequence of this approach, however,
the measure suffers from failing to distinguish between a normal economy of attention among actors sharing the same position
and stronger forms of polarisation. Moreover, because the measure focuses on the aggregate level of the communities as a whole,
it ignores both the different roles that actors adopt in the communities and their status. As Gould and Fernandez (1989) point
out in their seminal paper, not all actors can be expected to sustain similar relationships with other groups; indeed, the majority
of ties are usually managed by a restricted number of nodes, the
brokers, who thus occupy an important position in the network.
Differences in activity also mean that some brokers are more central
than others in the different communities. This should be particularly evident in open issues, where the ensuing polarisation further
segregates the communities from one another, which become more
introverted and increasingly depend on only a few actors to broker the relationships between them. In turn, because networks in
closed issues are more integrated, their communities should be
more extroverted, and the relationships between them should be
managed more evenly by a number of actors (Hypothesis 2).
Finally, our third hypothesis examines the degree of polarisation within the single clusters. In general, polarisation comes
into being when actors increasingly connect to like-minded others, that is, when network formation is strongly guided by the
homophily principle (McPherson et al., 2001). Clusters in polarised
networks are expected to closely follow this pattern: here, the ties
running between like-minded actors should be a factor that significantly contributes to the structure of the network. In turn, the
reverse should hold true for hyperlink networks in closed issues,
where homophily should play no significant role in the formation
of the network’s clusters, or it should explain significantly less than
expected (Hypothesis 3).
4. Approach and methods
4.2. Data collection: hyperlink networks
The present study uses a snowball technique, implemented
through web-crawling software (Issuecrawler, see Rogers, 2010),
to generate a climate change hyperlink network. As explained in
more detail below, the crawler starts from a set of specified seeds,
follows their outlinks, and thus generates a raw network that is
subsequently filtered with issue-specific keywords and reduced to
a co-link graph, whose actors are coded with their most important
attributes.3
The crawl starts from the most prominent German climate advocates and sceptics in terms of their ranking in country-specific
Google searches (see Adam et al., 2016, for a more technical description of the process).4 The number of seeds is restricted to four for
each side, as it proved difficult to find additional sceptics with a
prominence comparable to that of the advocates, due to the closed
nature of the issue. Introducing more seeds in pre-tests noticeably
increased the amount of noise but did not result in the inclusion of
a substantially higher amount of climate change actors in the core
of the network. Upon further examination of their level of engagement in the current debate and the hyperlink activity of the seeds,
we are confident that the selected seeds allow us to capture the
mainstream of the climate change issue as seen from a German
perspective. This is also evidenced by the relative stability of the
networks generated by several subsequent crawls each one month
apart.5 The seeds are civil society actors, as other actor types tend
to restrict their hyperlink communication to similar actors or their
own web domain (Petricek et al., 2006; Tremayne, 2005), which
the pre-tests confirmed. These tests also showed that the seeds
are well-connected actors in the resulting network as measured by
their average geodesic distance (see Appendix A in Supplementary
file), and together they cover the mainstream of the debate and
its central participants. We exclude digital platforms such as Facebook, for although they routinely become part of hyperlink crawls,
they are not themselves actors with their own positions—the precondition for investigating homophily effects (Fig. 1).
The crawler is set to stop adding new nodes after one iteration, that is, after having followed an outlink from a seed to its
target webpage, as crawls with more iterations result in an exponential growth of actors, most of which were not related to the
issue.6 As a last step, the crawler identifies the ties between the
4.1. Case selection: climate change in Germany
This study focuses on climate change as an issue that allows us
to test the hypotheses developed above, as it is an area of global
concern that transcends national boundaries and where actors can
take full advantage of the communicative potential of the internet. As a political debate, climate change fulfils the criteria for a
potentially fragmented and polarised issue because it pits those
who warn of the consequences of anthropogenic global warming
(‘advocates’) against those who see these concerns as unfounded or
exaggerated (‘sceptics’), with very little ground in between. From
the perspective of this study, climate change in Germany presents a
particularly interesting case since it conforms to our definition of a
closed issue. As one of the leading countries in the European Union
in terms of reducing greenhouse gases, German climate change politics are driven by a broad alliance of advocates in a hegemonic
position, including actors from the government and administration,
the parliamentary parties, research institutions, and environmental
movement organisations (Engels et al., 2013; Jost and Jacob, 2004;
Weidner and Mez, 2008). Additionally, scepticism is marginalised
in both public opinion, where only 2% of the population think
that climate change is ‘not a serious problem at all’ (European
Commission, 2011, p. 72), and in media coverage, with German
newspapers reporting nine times fewer sceptical voices than in the
United States (Grundmann and Scott, 2014).
3
An alternative approach would consist of collecting all webpages of interest via
appropriate search queries, for instance, with Google, and then determine the links
that run between them. This is above all a viable method when the issue under
study is clearly demarcated by search terms—possibly neologisms—and results in a
manageable population. Potential problems might arise with issues such as climate
change, which is grounded in the national political space but has strong transnational dimensions. From the perspective of the present study, transnational actors
should not be excluded a priori, but neither should they be given too much weight
in the first step of the sampling process, as this would dilute the national focus and
hence the research interest. Given, furthermore, that the shift between the national
and transnational spaces in this case coincides with a change of language, the present
study opted for a link-tracing design, which allows transnational actors to become
part of the network but places a clear emphasis on national actors.
4
The search terms used, as well as the source seeds, are given in Appendix A in
Supplementary file.
5
A drawback of this approach and the co-link method used to reduce the graph
is to be seen in the fact that we miss the actors at the margins of the debate and the
more isolated voices. In other words, the methods used in this study are expressly
designed to map the mainstream of a debate. A study interested in those actors
that are more marginal in an issue would have to choose different procedures, for
instance, by selecting appropriate seeds for a link-tracing procedure from queries in
search engines that specifically concentrate on the lower-ranked actors. Tracing the
connections between these actors directly or feeding them as seeds in corresponding
web-crawls could then help to capture that part of the issue population that resides
at the periphery of the debate and is only marginally connected to the mainstream.
6
This is confirmed by two recent crawls (November 2016) based on the same
seeds and conducted with one and two iterations, respectively. They show that
Please cite this article in press as: Häussler, T., Heating up the debate? Measuring fragmentation and polarisation in a German climate
change hyperlink network. Soc. Netw. (2017), https://doi.org/10.1016/j.socnet.2017.10.002
G Model
SON-1053; No. of Pages 11
ARTICLE IN PRESS
T. Häussler / Social Networks xxx (2017) xxx–xxx
5
Fig. 1. Steps to generate the German climate change hyperlink issue network.
actors that are now part of the network. There is also a theoretically
based reason for having a rather restrictive boundary specification:
because climate change is an issue of global concern, it involves
actors from other countries that are situated in different issue contexts. The research design attempts to balance the tension between
the theoretical interest, which is related to country-specific issue
configurations of public debates as the political context, and the
practical view that one of the central affordances of online communication consists of expanding relationships beyond political,
social, and geographical borders. As a result of these considerations,
the network generation procedure employed in this research gives
preference to the linking behaviour of German actors, in particular
the seeds, whose hyperlink patterns constrain the relationships of
actors from other countries.
Because the crawler follows all of the detected hyperlinks
regardless of the actors’ type, position, or country (Gonzalez-Bailon,
2009), the network contains a substantial amount of noise in terms
of actors and content that is irrelevant to the issue. All the webpages are therefore indexed with a set of issue-specific keywords
in German and English, discarding all those that display no match,
so that the resulting network contains only actors addressing climate change and who attribute topical relevance to other actors
through their hyperlinks.
The nodes are filtered to have at least two links, one of which
is an inlink, to guarantee a minimum of recognition by another
actor in the issue. This step reduces the number of actors from
415 (original crawled and indexed network) to 244 (co-linked network) and has the additional advantage of making the network
more amenable to inferential analyses (Fig. 2). As a last step, all of
the actors in the network are classified manually by two coders
according to their type, country of activity, and position on the
issue.7
the wider the boundary specified, the lower the share of the issue in the network:
whereas in the network based on one iteration 41% of all webpages (4,696 of 11,578)
dealt with climate change as an issue, only 17% did so for the crawl with two iterations (31,117 webpages of a total of 179,247). On the other hand, the more the
boundary is extended, the less relevant the political context of the seeds becomes
for how the network is shaped. In the network with a boundary of one iteration,
60% of the webpages belong to German actors (6,945 of 11,578), while in the larger
network the value reaches only 28% (7,575 of 31,117). The restrictive boundary
specification thus guards against losing both the issue and the political context.
7
The crawl was conducted in June 2012. Websites that belong to the same actor
are grouped together. The actor attributes are coded according to how they present
themselves on their websites. The country of activity is preferred to the country
of origin, as the two do not necessarily coincide, above all for nongovernmental
organisations (NGOs). Additionally, the actors are classified according to type, distinguishing between political actors (governments, their agencies, political parties,
etc.); actors from the economic sector including business associations; media (both
traditional and online media); actors from civil society including NGOs, think tanks,
research institutions, etc.; and nonaffiliated private persons, mostly bloggers. The
actor’s position is taken from the information on their web pages, mostly from the
‘about’ section. An exception is online media outlets, where up to three editorial articles were coded to determine, if possible, the stance of the organisation. The actors
are classified as climate advocates, sceptics, ambivalent, or as having no position.
Fig. 2. Communities in the German climate change hyperlink network. (For interpretation of the references to colour in this figure legend, the reader is referred to
the web version of this article.)
Note. Node size is proportional to indegree.
Blue = C1 community.
Red = C2 community.
Green = C3 community.
Black = Marginal communities.
4.3. Data analysis: measures of polarisation
Hypothesis 1 addresses the general level of the network and
postulates that although hyperlink networks in closed issues might
fragment into smaller communities, the distribution of the actor’s
positions should not differ significantly from each other. To test
this hypothesis, we will first identify neighbourhoods of wellconnected actors in the network through an inductive community
detection algorithm. Then, we will use a Chi-square test to determine whether the distribution of the positions differs significantly
between the clusters. The inductive community detection approach
is similar to that employed by Elgesem et al. (2015) and Williams
et al. (2015) and has the advantage that it guards against the economy of attention fallacy pointed out by Benkler (2006), as clusters
are formed purely on the grounds of the actors’ linking patterns,
regardless of their position on the issue, country, or type.
Hypothesis 2 tests whether fragmentation and polarisation go
hand-in-hand by examining the connections between the clusters.
In closed issues, communities and their brokers should display an
overall extrovert orientation towards other communities, and the
ties between them should be evenly moderated by several actors.
Guerra et al. (2013) introduce an aggregate polarisation measure P
for network communities that captures the extent to which brokers
from one community—the ‘boundary’ B in their terminology—on
average connect to those of another community (di (v)) in relation to the ties they sustain to members of their own community
(db (v)) and the boundary connections (see the formal definition
in the equation below). Subtracting 0.5 from this value and averaging the result over all members of the boundary yields a figure
that lies in the range [-0.5, +0.5]. Negative values indicate a higher
degree of broker connectivity towards those of other communities and can be taken as a sign of network integration, whereas
Inter-rater reliability is based on a comparison of 150 coded units for each variable
of the two coders’ ratings with a master template (Krippendorff, 2012;). The coefficients can be deemed satisfactory: using Krippendorff’s alpha, the mean values are
0.898 (actor type), 0.927 (country), and 0.891 (actor position).
Please cite this article in press as: Häussler, T., Heating up the debate? Measuring fragmentation and polarisation in a German climate
change hyperlink network. Soc. Netw. (2017), https://doi.org/10.1016/j.socnet.2017.10.002
G Model
ARTICLE IN PRESS
SON-1053; No. of Pages 11
T. Häussler / Social Networks xxx (2017) xxx–xxx
6
positive values imply a more introvert orientation and suggest tendencies of network fragmentation. While Guerra et al. (2013) offer
no test of statistical significance such as a permutation test, Pvalues close to zero can generally be interpreted as representing
normal economies of attention, whereas those more to the positive extreme of the scale signify that the community relationships
are marked by a substantial amount of polarisation; large negative
values in turn indicate network integration.
P=
di (v)
1
− 0.5
B
db (v) + di (v)
v∈B
Guerra et al.’s (2013) measure has the advantage of providing an
aggregate view of community interaction while concentrating on
those actors responsible for the connectivity between them—the
brokers. They thus make an important distinction in the roles of
the community members, which is often lacking in the hyperlink
polarisation literature.
To assess the role of single brokers, we will use Gould and
Fernandez’s (1989) typology for directed networks and concentrate on the two most important roles for this research: those
that connect actors from within their own clusters to the outside,
the ‘representatives’, and those that mediate links in the opposite
direction, the ‘gatekeepers’. We will then examine whether overall
representative ties occur more frequently than gatekeeper connections and test whether brokerage roles and status are evenly
distributed between the actors.
Finally, Hypothesis 3 contends that while polarisation should be
visible in open issues on the local level of tie formation processes,
in terms of actors connecting to like-minded others, it should play
no significant role in closed ones. To examine the degree to which
homophily is a significant factor in shaping the communities, we
will use an inferential approach. ERGMs are a family of statistical models that allow us to examine the interdependencies of
actors and test whether their local choices are significantly driven
by specific actor attributes—such as positional homophily—or
whether other factors better explain the network formation process
(Cranmer and Desmarais, 2011; Gonzalez-Bailon, 2009; Goodreau
et al., 2008). We will test the homophily effect of the actor’s position, country, and the actor type on network formation for each of
the identified communities.
5. Findings
We first present the structural properties and the nodal
attributes for both the network as a whole and its communities. The
three clusters listed in Table 1 below were identified in an inductive way using a community detection algorithm, in this case the
‘walktrap’ algorithm (Pons and Latapy, 2005) of the igraph package
in R (Csardi and Nepusz, 2006). The working logic of this and similar
algorithms is based on random walks in the network, which identify
denser areas that can then be extracted as isolated subgraphs. Other
popular community detection algorithms are based on betweenness, such as the Newman–Girvan algorithm (Newman and Girvan,
2004), which proceeds hierarchically by iteratively deleting the
edge with the highest betweenness and then calculating the modularity of the resulting graph, or they follow an agglomerative
procedure, such as the ‘fastgreedy’ algorithm, which proceeds in
the opposite way by merging those nodes that optimise the modularity of a community (Clauset et al., 2004). These algorithms tend
to produce different results, and the present study uses the ‘walktrap’ algorithm because it allocated the greatest number of nodes
in the fewest number of communities, although the changes in the
results yielded by the various algorithmic solutions were rather
modest. Furthermore, the ‘walktrap’ algorithm has been shown to
produce rather conservative results (Li et al., 2017).
Table 1
Frequencies of general network measures and actor attributes.
Frequencies
Community1 Community2 Community3 Total network
Number of nodes
Number of edges
Average degree
Diameter
Average distance
Density
Actor type
Political
Economic
Media
Civil society
Private person
Other/n.a.
Actor country
Germany
US
UK
Transnational
Other/n.a.
Actor position
Advocate
Sceptic
Other/n.a.
98
246
2.5
5
2.2
0.026
87
356
4.1
9
3.2
0.048
26
38
1.5
2
1.2
0.058
244
885
3.5
11
5
0.014
10
3
47
25
11
2
6
7
23
20
30
1
4
0
17
4
1
0
32
11
99
54
45
3
32
18
6
30
12
13
40
9
12
13
21
0
3
1
1
76
62
18
54
34
87
2
9
27
47
13
24
0
2
167
49
28
Note. The figures for the total network represent the aggregate values for all of the
identified communities and therefore differ from the sum of the three clusters.
In the present case, the ‘walktrap’ procedure detected three
larger components, which combine over 87% of the nodes, and several smaller communities, often containing only one or two nodes,
which were excluded from further analysis.
Table 1 lists several network measures as well as the frequencies for the attributes of the actors. On this general level the basic
network conforms to our expectations about an online space representing a closed issue, as climate advocates occupy a hegemonic
position, accounting for 68% of the actors, whereas a minority of
only 20% are climate sceptics. The network also embodies a strong
transnational orientation: less than one-third of the actors are from
Germany, closely followed by actors from the United States and
transnational actors, who consist mostly of supranational bodies
such as the United Nations Framework on Climate Change as well
as a few global media websites. With regard to the actor types,
the network displays a diverse pattern. The combined categories
of civil society organisations and private persons (the blogosphere)
account for the same number of nodes as the media, which indicates
that climate change is not only a debate between actors holding
antagonistic positions but also a struggle for interpretive authority and dominance in the public sphere. Compared to this, political
actors are in the minority while those from the economy have only
a marginal presence.
Turning to the communities, we can see that they differ not only
in size but also in their composition. Of the three communities,
two (C1 and C3) are clearly dominated by advocates, whereas the
C2 fragment is the only cluster with a sceptical majority—in fact,
it comprises nearly all of the sceptics in the network—although it
also contains a substantial number of advocates.
5.1. Network fragmentation
Having examined the general characteristics of the network and
its main communities, we can now proceed to test the hypotheses. Our first hypothesis states that although hyperlink networks
in closed issues might fragment into smaller clusters, the distribution of the actors’ positions in these clusters should not differ
significantly from each other. In other words, instead of uniformity we should find communities embodying similar degrees of
diversity. Comparing the distributions of the actors’ positions in
Please cite this article in press as: Häussler, T., Heating up the debate? Measuring fragmentation and polarisation in a German climate
change hyperlink network. Soc. Netw. (2017), https://doi.org/10.1016/j.socnet.2017.10.002
G Model
ARTICLE IN PRESS
SON-1053; No. of Pages 11
T. Häussler / Social Networks xxx (2017) xxx–xxx
7
Table 2
Chi-square test of position distributions between the communities.
Test
X-squared
Degrees of freedom
p-value
C1 , C2 , C3 (N = 211)
88.9844
n.a.
0.0005
Note. Given the low values for sceptics in C1 and C3 , the test uses a simulation method
based on 2000 replicates. An additional Fisher exact test yielded an even lower
p-value (p < 2.2e-16).
the clusters employing a Chi-square test, we can see in Table 2 that,
contrary to our assumption, the distributions of the three clusters
differ significantly from each other.
Although taken on its own the mixed C2 cluster conforms more
strongly to our expectations of an integrated online space, overall
the actor distributions between the three communities provide a
clear sign of polarisation, which leads us to refute the first hypothesis. If polarisation is also at work in closed issues, however, this
should become apparent in the ties between the communities and
the role played by brokers, which is what the second hypothesis
assesses in more detail.
5.2. Community brokerage
The second hypothesis examines the external and internal relationships of the boundaries of the communities (Guerra et al., 2013),
and in a second step focuses more specifically on those actors
who broker the majority of the ties between the clusters, thereby
concentrating on those that either act as ‘representatives’ (inside
to outside) or as ‘gatekeepers’ (outside to inside) in Gould and
Fernandez’s (1989) classification. Several aspects of these constellations are of interest in our context: first and very generally, we are
interested in the extent to which the single communities on average demonstrate an extrovert (depolarised) or introvert (polarised)
orientation. Second, building on this approach and taking a more
detailed view, we assess the number of brokers acting in the two
roles, and the number and direction of ties they mediate between
the inside and the outside of the communities; here, we concentrate
on those actors that together broker at least 90% of the relationships. Third, we examine their characteristics in terms of where
they stand on the issue, their country of activity and actor type.
Finally, we test whether the number of ties the actors broker is
above the threshold of statistical significance. In closed issues, relationships between different communities should be mediated to
similar degrees by several actors and the observed values should
not differ greatly from the expected ones. The table below presents
the P value (Guerra et al., 2013) for the general orientation of the
communities towards each other along with the observed brokerage values (Gould and Fernandez, 1989), as well as the standardised
values for those actors in the three communities that together
mediate more than 90% of all the relationships between the inside
and the outside of the single clusters (Table 3).
The first thing to note is that the brokers of two of the three
communities (C1 and C2) display a P value that indicates a low
to moderate degree of polarisation, and is most pronounced in
the politically heterogeneous and US-dominated community C2,
whereas the brokers of the smallest community, C3, are neither
particularly extroverted nor introverted.8
As far as the roles and status of single nodes are concerned, the
majority of ties (over 90%) are brokered by a minority of nodes
(between 4% and 8%); at the same time, there are obvious differ-
8
Guerra et al. (2013) develop their measure in the context of undirected networks,
which has the disadvantage that we are not able to tell who is responsible for the
resulting integration/fragmentation processes. Adapting their measure to directed
networks yielded results that confirm the tendency of the original P values.
Fig. 3. Issue positions in the German climate change hyperlink network. (For interpretation of the references to colour in this figure legend, the reader is referred to
the web version of this article.)
Note. Node size is proportional to indegree.
Red = Climate sceptic.
Green = Climate advocate.
Blue = No position/ambivalent.
ences in the terms of brokerage activity and orientation, which only
become apparent when examining the directed network. In this
view, the C1 community is more extroverted in that its actors moderate more relationships from the inside to the outside, whereas the
opposite is the case for the other two clusters, which are dominated
by gatekeeper relationships and display a more introspective orientation, indicating a greater degree of polarisation (see Guerra et al.,
2013). The two larger communities are furthermore characterised
by a substantive number of transnational actors as important brokers. This is above all true in the mixed community C2, where
actors from the United States, where climate change is very actively
debated, occupy central roles, although they occupy different positions on the issue. Almost half of the representative relationships
are brokered by Watts Up With That, a prominent climate sceptic
blog, and 36% of the ingoing links are moderated by RealClimate, an
advocate website run by climate scientists. By contrast, the brokerage roles in the smallest cluster, C3, are occupied by a more uniform
set of actors, which consists entirely of German climate advocates,
mainly political institutions (ministries and federal offices), and
research institutes. The relationships that the C1 cluster establishes
with the other communities in turn depend strongly on one actor,
the Heinrich Boell Stiftung, who reaches the highest values in both
brokerage roles.
The analysis thus yields a somewhat mixed picture, as at the
aggregate level two of three communities appear to be more
introverted than expected. This view is partly confirmed but also
rendered more complex when examining the role and status of single brokers in the directed network. Here, network integration and
diversity in terms of the ‘representative’ brokers in the C1 community and those from across the political divide in the C2 cluster
are contrasted with the segregationist tendencies demonstrated by
the dominance of a select few actors who mediate the vast number of the relationships. Overall, and similar to the top level of
the network, the analysis of the intermediate level reveals unexpected polarisation effects that govern the relationships between
the communities, and we therefore refute our second hypothesis
(Fig. 3).
Please cite this article in press as: Häussler, T., Heating up the debate? Measuring fragmentation and polarisation in a German climate
change hyperlink network. Soc. Netw. (2017), https://doi.org/10.1016/j.socnet.2017.10.002
G Model
SON-1053; No. of Pages 11
ARTICLE IN PRESS
T. Häussler / Social Networks xxx (2017) xxx–xxx
8
Table 3
Actor brokerage scores for 90% of mediated ties by community and total P-values.
Community
C1
Total
P value
C2
Total
P value
C3
Total
P value
Actor
Country
Position
Representative
Gatekeeper
Observed
Observed
Standard.
Standard.
Center for American Progress Action Fund
Greenpeace International
Heinrich-Böll-Stiftung
Klimaretter.info
The Guardian
UNFCCC
WWF Germany
US
Global
DE
DE
UK
Global
DE
A
A
A
A
A
A
A
0
0.89
13*
2.40
258*
63.99
37*
8.43
15*
2.90
0
−0.87
0
−0.87
330 (N 339)
12*
2.15
7
0.89
89*
21.51
0
−0.87
10
1.65
7
0.89
6
0.64
131 (N 146)
0.16
Climategate.nl
Die kalte Sonne
EIKE − Europäisches Institut für Klima und Energie
International Climate and Environmental Change Assessment Project (icecap)
JoNova
RealClimate
The New York Times
Watts Up With That?
NL
DE
DE
US
AU
US
US
US
S
S
S
S
S
A
A
S
13*
2.46
20*
4.24
0
−0.85
19*
3.99
11*
1.95
28*
6.28
18*
3.73
135*
33.55
243 (N 273)
0
−0.85
0
−0.85
47*
11.12
0
−0.85
0
−0.85
174*
43.49
34*
7.81
77*
18.77
332 (N 361)
0.24
Bundesministerium für Bildung und Forschung
Bundesministerium für Umwelt, Naturschutz und Reaktorsicherheit
Max-Planck-Institut für Meteorologie
Potsdam Institute for Climate Impact Research
Technische Universität Berlin
Umweltbundesamt
Weltenwetter
DE
DE
DE
DE
DE
DE
DE
A
A
A
A
A
A
A
3
1
2
0
0
7*
0
13
1
0
0
20*
1
0
4
26
0.60
−0.18
0.21
−0.56
−0.56
2.15
−0.56
−0.18
−0.56
−0.56
7.18
−0.18
−0.56
0.98
0.03
Note. Standardized z-values were computed by subtracting the expected value from the observed value and dividing the difference by the standard deviation of the expected
value (Gould & Fernandez, 1989; p. 113ff.). Values above 1.96 can be considered to be statistically significant; the corresponding observed values are marked with an asterisk
(*).
The P values were computed according to the definition in Guerra et al. (2013) for all brokers in the communities.
Abbreviations. Country: AU = Australia, DE = Germany, NL = Netherlands, US = United States.
Position: A = Climate change advocate, S = Climate change sceptic.
5.3. Tie formation and homophily
So far, our analysis has produced ambiguous results, as the network and its communities have revealed signs of both polarisation
and integration. In this context, our third hypothesis sheds more
light on the internal communication processes that shape the single communities and by extension the issue as a whole. Using
ERGMs allows us to determine the extent to which the tie formation processes in the communities are driven by homophily; that
is, by the actors’ orientation towards like-minded others. ERGMs
have an advantage over logistic regression models because they
are explicitly designed to estimate relational data, which form the
very basis of social networks, and which include dyadic as well as
indirect relationships between three or more actors (Cranmer and
Desmarais, 2011; p. 66). In social networks, the presence of an actor
depends on the presence of one, or indeed several other, actors, and
ERG models thus overcome the problems of traditional statistical
approaches, where the observations are taken to be independent
from one another (Lusher et al., 2012, p. 12ff.). More specifically,
ERGMs estimate the configuration of a network based on specific
structural parameters, nodal attributes such as homophily effects,
and exogenous factors such as other networks that are of theoretical interest (Lusher et al., 2012; p. 12). The basic assumption is
that the observed network comes from a random distribution, and
the ERG process estimates the extent to which the specified model
significantly differs from these other networks and is thus able to
capture its distinct structure. This allows us to arrive at a better
understanding of the factors underlying the nodes’ tie formation.
In the present case, each community represents a separate network, and we therefore specify separate models for each. As we are
interested in the degree to which positional homophily affects the
generation of network ties, the structural parameters of the model
primarily act as covariates that guard us against overestimating
the effect of relationships between like-minded actors (GonzalezBailon, 2009; p. 277), although they are also important factors in
themselves, for as we will see they are able to reveal the social
processes and hierarchies that characterise the single communities. The parameters differ between the models, as they capture
the specific structure of each network. The table below presents
the results of the ERGMs fitted to the three communities.9 The estimates are measured on a logit scale and can be interpreted simply
as unstandardised coefficients in logistic regression (Table 4).
The upper half of the table consists of the structural factors, and
the lower half reports the coefficients for the nodal attributes. Most
of the structural coefficients have a negative sign in the larger communities, which means that they occur at a rate that is lower than
expected, a common feature of graphs where the nodes are only
sparsely connected. Because the C1 and the C3 communities consist
almost exclusively of climate advocates, the relationships between
like-minded actors in these cases should not be assessed at the
level of the position of the actors as the attribute lacks the necessary variance; rather, it is important to take into account whether
the structural parameters that model the network suggest an even
and symmetric connective pattern.
The lack of ideological and geographic homophily is particularly evident in the C3 community, where neither the position of
9
The ERGMs were fitted using the ergm package in R (Handcock et al., 2016;
Hunter et al., 2008). The exact specifications of the structural model terms are given
in (Morris et al., 2008). The goodness of fit procedures for the estimated models that
returned acceptable to good values, and the corresponding plots, can be found in
Appendix B in Supplementary file.
Please cite this article in press as: Häussler, T., Heating up the debate? Measuring fragmentation and polarisation in a German climate
change hyperlink network. Soc. Netw. (2017), https://doi.org/10.1016/j.socnet.2017.10.002
G Model
ARTICLE IN PRESS
SON-1053; No. of Pages 11
T. Häussler / Social Networks xxx (2017) xxx–xxx
9
Table 4
ERGMs of the communities C1 , C2 , and C3.
C1 (N = 98)
Structural parameters
Edges
Gwesp
Gewsp alpha
Asymmetric
Triad 021U
Triad 021C
Triad 102
Intransitive
Threepath RRL
Threepath LRL
Twopath
Indegree sqrt
Outdegree sqrt
Node level attributes
Position
A-A
A-S
Country
DE-DE
US-US
US-DE
Actor type
Pol.-Pol.
C2 (N = 87)
C3 (N = 26)
Estimate
Std. Error
Estimate
Std. Error
Estimate
Std. Error
−9.21***
–
–
–
–
–
0.03***
−0.34***
−0.02***
−0.01***
0.36***
2.05***
1.20***
0.33
–
–
–
–
–
0.01
0.09
0.01
0.001
0.09
0.14
0.08
−10.11***
0.25†
−0.59
−0.79**
−0.24***
0.10***
–
−0.31***
−0.01***
−0.01***
0.22***
3.18***
1.34***
0.59
0.13
0.41
0.28
0.05
0.02
–
0.04
0.002
0.001
0.04
0.25
0.07
−9.50***
–
–
–
–
–
–
–
–
–
–
3.18***
1.75***
1.22
–
–
–
–
–
–
–
–
–
–
0.68
0.21
–
−3.18*
–
1.48
0.63**
−0.87**
0.24
0.29
–
–
–
–
1.33***
1.89***
–
0.23
0.33
–
1.42***
0.37†
−1.92**
0.27
0.19
0.76
–
–
–
–
–
–
1.55***
0.41
–
–
–
–
Note: p-values: < 0.1 = † , < 0.05 = *, < 0.01 = **, < 0.001 = ***
Abbreviations. Position: A = Climate advocate, S = Climate sceptic. Country: DE = Germany, US = United States. Actor type: Pol. = Political actor.
the actors nor their country or type has a significant effect on the
formation of the ties and thus the shape of the cluster. Although
the community consists almost entirely of climate advocates, the
tie generation process is strongly influenced by two purely structural parameters, the sociability and above all the popularity of
some nodes. These effects reveal an additional process at work
beyond those suggested by the first two hypotheses, for although
the community is undoubtedly formed by the interaction between
like-minded actors, this explains its structure only in part. As we
can see, at the local level, activity and attention are not evenly distributed, and the resulting social stratification is due to the nodes’
preference to form ties with those commanding a high prestige
(popularity) rather than to connect to any node sharing the same
position. This factor is complemented by the higher tie formation
patterns of some nodes (sociability), and together they are able to
explain the structure of this ideologically homogeneous network
community.
Similar dynamics are also the dominant characteristic of the C1
community, which is composed almost exclusively of climate advocates, complemented by those structural configurations that model
indirect, mostly unreciprocated connections between three or four
actors. In this community, however, these endogenous factors are
accompanied by two homophily effects: the preference of political
actors to connect to each other and the geographic preference of
German and US actors to link to other actors from the same country. Actors in the C2 community display a similar trend with regard
to geographic homophily. Although the hyperlink crawl starts from
German seeds and subsequently moves to actors from the United
States, in both the C1 and the C2 communities the transatlantic
relationships are established only in single instances and are on the
whole too weak to produce a significant pattern; the actors mainly
connect to those from the same country and language area. At the
same time, the C2 cluster is the only community where we can find
a positional homophily effect, with advocates linking to each other
to a degree higher than expected. By exponentiating the positional
homophily coefficient (0.63) we get the odds ratio (1.88) of the tie
generation pattern, which in this case means that a climate advocate forming a new tie is almost twice as likely to connect to another
advocate than to a sceptic or an actor with no position. This result is
perhaps surprising, given that the cluster contains the largest number of climate sceptics, but it appears that their relationships are
not based on associations with like-minded others.
Connections across the political divide in turn are lower than
would be expected both for the C1 and the C2 communities, which
is particularly relevant in the latter case since this cluster has a substantial number of members from both camps (50% sceptics, 31%
advocates). Indeed, here the effect is a partial internal fragmentation and polarisation of advocates, which is exacerbated by the fact
that more integrative linking across the camps of the debate either
is non-significant (sceptics to advocates) or occurs to a significantly
lesser extent (advocates to sceptics).
Taken together, these results largely lead us to reject our third
hypothesis, and the analysis reveals the multi-layered nature of
network formation processes. In the two largest communities, the
main homophily effect is given by the actors’ country of activity, whereas positional homophily plays only a minor role in all
three clusters. This is due to a lack of variance in the positional
attribute in the case of the first and the last community; in the
C2 cluster, however, the story is a different one, and the results
are rather unexpected. Contrasting some of the findings of other
studies, which have documented communities that comprise actors
with opposing views (Elgesem et al., 2015; Williams et al., 2015),
the results here show that the open forum character of the community gives way to an internally fragmented space, where little
significant interaction occurs between climate advocates and sceptics. At the same time, we can also see that the political effect of
connections between like-minded actors is interrelated with the
effects of status and prestige.
6. Discussion and conclusion
The present study has contributed to developing a more detailed
methodology that allows us to distinguish different forms of
polarisation and measure their relative strength at the general,
intermediate, and local levels of a network. As the analysis has
shown, we need all three levels to arrive at a full understand-
Please cite this article in press as: Häussler, T., Heating up the debate? Measuring fragmentation and polarisation in a German climate
change hyperlink network. Soc. Netw. (2017), https://doi.org/10.1016/j.socnet.2017.10.002
G Model
SON-1053; No. of Pages 11
10
ARTICLE IN PRESS
T. Häussler / Social Networks xxx (2017) xxx–xxx
ing of what degree, and in what way, an issue network is shaped
by various forms of homophily. At the same time, the study has
extended the focus from those settings that are ex ante assumed
to be polarised to those issue constellations that are marked by
lower levels of controversy and have received scant attention so
far. While based on the theoretical account we would expect a
largely depolarised political online space, the results show that the
inherent issue dynamics are not as straightforward, as the closed
issue examined demonstrates complex fragmentation tendencies
accompanied by significant polarisation effects. The lower degree of
contentiousness of an issue offline is thus no guarantee for finding
the same constellation online.
Accordingly, the German climate change network already
displays a certain degree of polarisation at the top level of
the issue space, which fragments into three communities, two
of which are mostly composed of like-minded actors—climate
advocates—whereas the mixed third one comprises nearly all the
sceptics. These results support the general view that polarisation results from a sorting process, which leads to ideologically
homogenous communities. The question then is how strongly these
communities are connected to each other.
The results show that polarisation is also evident in the relationships between the communities, although here we already see
a more intricate picture emerging. On the aggregate level, the communities are characterised by largely absent (C3) and low (C1) to
moderate (C2) degrees of polarisation, which cautions us against
jumping to conclusions based on the analysis of the top level of the
network alone. The intermediate network level view becomes yet
more complex once we examine the roles of single brokers in the
directed networks. While the ties between the inside and the outside of the largest advocate community (C1) are managed by only
a few actors—and indeed depend on one very influential broker—it
is also the community where the role of the ‘representatives’ outweighs that of the ‘gatekeepers’—thus indicating integration rather
than segregation. The heterogeneous C2 community in turn has
actors from both camps acting as important brokers, whereas the
smallest cluster C3 is characterised by an uneven distribution of
brokerage patterns across the nodes, although on the whole the
‘gatekeeping’ function is more pronounced and managed mainly by
one actor. Regardless of their particular connective orientation, all
the communities depend on a few brokers, who manage the relationships between the inside and the outside, and while some of
them help to integrate the network across community boundaries,
small changes in their orientation can have profound consequences
for the network structure. In other words, the potential of the communities to become further polarised should not only be assessed
on the level of their aggregate relationships but also in terms of
how the connective activity is stratified across the brokers, which
in the present case is concentrated in the activity of a few nodes.
Finally, the analysis shows that on the local level of tie formation
processes, ideological homophily is not a decisive factor in the C1
and C3 communities, although this is largely due to their positionally homogenous populations. In fact, positional homophily only
plays a partial role in the mixed C2 cluster, where advocates primarily form ties with other advocates, whereas the community’s
sceptical majority displays no clear connective preferences. This
result shows the importance of including the local level in the
analysis, as otherwise we might conclude that this community represents an open forum, where both sides intensely interact with
each other.
The more we move from the level of the composition of the communities and their relationships—the levels most often analysed in
the existing literature—to the local level of the single nodes’ tie
formation process, the less political segregation alone is decisive
in explaining the network structure, as other, social, factors prove
to be more dominant. Polarisation in the sense of relationships
between like-minded actors does not necessarily mean that they
are all equally connected. As we have seen, the apparent uniformity of the C1 and C3 communities at the top level of the network
takes on more complex forms at deeper levels, which are marked
by social stratification, and sparse and one-sided relationships, as
revealed by both the brokerage analysis and particularly the ERG
models.
Despite the geographic homophily effects documented by the
ERGMs, the overall dynamics at work in the German climate change
discourse are significantly affected by the transnational orientation
of German actors, which leads to the inclusion of supranational
institutions, but in particular of US advocates and sceptics in the
network. Of course, the specific effect generated by this constellation requires further examination, but it suggests that actors in
low-contentious issues tend to connect to those in highly contentious contexts (Häussler, et al., 2017). This could be due to the
fact that higher levels of controversy generate new arguments and
information, which easily diffuse through hyperlink spaces and
thus enrich the debates in closed issues. German climate sceptics
in particular, due to their weak position in the debate and inability
to mobilise a critical mass at the national level, have an inherent
interest to forge connections to those contexts where their side
of the issue is more established (see Keck and Sikkink, 1999; for
the relevance of transnational advocacy networks). At the same
time, the ERG models show that these relationships occur only
in single instances and do not lead to an integrated transnational
sceptical community. Furthermore, because of their transnational
orientation, German climate sceptics reside in a cluster that is
largely disconnected from the domestic debate and where connections to the other communities are mainly managed by US
sceptics. Whether the existing connections are nevertheless apt to
strengthen the position of German climate sceptics in the national
debate, above all their visibility on the agendas of the public, the
media, and politics is an open question and an aspect that should
be examined in longitudinal studies.
In contrast to the German offline context, which is clearly
depolarised, the corresponding network connects low- and highcontentious settings, resulting in a hybridised political structure.
The results of this study suggest that we need a more fine-grained
approach that allows us to distinguish different configurations and
stages of the political process—including in particular its transnational dimension—and future research should examine how linking
patterns change when issues become more contentious as well as
how this affects polarisation at the different levels of the network.
The brokerage perspective has allowed us to identify specific nodes
that manage a substantial number of the relationships between the
communities, and their role in the evolution of issues networks
should receive particular attention.
This brings us to a final point. While it can be argued in the
present case that there are good scientific reasons why the dissenting sceptical voices in political debates should be marginalised,
other issues, above all those with strong moral dimensions,
face different theoretical and empirical questions. Many of the
democratising forces of the past decades and centuries started
as oppositional counter-movements that challenged the existing
hegemony of the status quo, generating substantial controversy
and contributing to a polarisation of the political space. We
therefore also need a better theoretical and methodological understanding of the different forms of polarisation that allows us to
distinguish between the beneficial and detrimental consequences
of fragmentation dynamics and homophily effects for the political
process and the public sphere.
Please cite this article in press as: Häussler, T., Heating up the debate? Measuring fragmentation and polarisation in a German climate
change hyperlink network. Soc. Netw. (2017), https://doi.org/10.1016/j.socnet.2017.10.002
G Model
SON-1053; No. of Pages 11
ARTICLE IN PRESS
T. Häussler / Social Networks xxx (2017) xxx–xxx
Acknowledgement
This study was funded by the Swiss National Science Foundation
under grant no. 100017E-135915.
Appendix A. Supplementary data
Supplementary data associated with this article can be found,
in the online version, at https://doi.org/10.1016/j.socnet.2017.10.
002.
References
Adamic, L., Glance, N., 2005. The political blogosphere and the 2004 US election:
divided they blog. Paper Presented at the Proceedings of the 3rd International
Workshop on Link Discovery.
Adam, S., Häussler, T., Schmid-Petri, H., Reber, U., 2016. Identifying and analyzing
hyperlink issue networks. In: Vowe, G., Henn, P. (Eds.), Political
communication in the online world. Theoretical approaches and research
designs. Routledge, New York, pp. 233–247.
Adamic, L., 2008. The social hyperlink. In: Turow, J., Tsui, L. (Eds.), The Hyperlinked
Society. University of Michigan Press, Ann Arbor, MI, pp. 227–249.
Baumgartner, F.R., Jones, B.D., 1991. Agenda dynamics and policy subsystems. J.
Polit. 53 (04), 1044–1074.
Benkler, Y., Roberts, H., Faris, R., Solow-Niederman, A., Etling, B., 2015. Social
mobilization and the networked public sphere: mapping the SOPA-PIPA
debate. Polit. Commun., 1–31 (ahead-of-print).
Benkler, Y., 2006. The Wealth of Networks: How Social Production Transforms
Markets and Freedom. Yale University Press, New Haven.
Bimber, B., 2003. Information and American Democracy: Technology in the
Evolution of Political Power. Cambridge University Press, Cambridge.
Chaffee, S.H., Metzger, M.J., 2001. The end of mass communication? Mass
Commun. Soc. 4 (4), 365–379.
Clauset, A., Newman, M.E., Moore, C., 2004. Finding community structure in very
large networks. Phys. Rev. E 70 (6), 1–6.
Converse, P.E., 1987. Changing conceptions of public opinion in the political
process. Public Opin. Q. 51, 12–24.
Cranmer, S.J., Desmarais, B.A., 2011. Inferential network analysis with exponential
random graph models. Polit. Anal. 19 (1), 66–86.
Csardi, G., Nepusz, T., 2006. The igraph software package for complex network
research. Int. J. Complex Syst. 1695 (5), 1–9.
Elgesem, D., Steskal, L., Diakopoulos, N., 2015. Structure and content of the
discourse on climate change in the blogosphere: the big picture. Environ.
Commun. 9 (2), 169–188.
Engels, A., Hüther, O., Schäfer, M., Held, H., 2013. Public climate-change skepticism,
energy preferences and political participation. Glob. Environ Change 23 (5),
1018–1027.
European Commission (2011). SpecialEurobarometer 372: Climate change.
Retrieved from http://ec.europa.eu/public opinion/archives/ebs/ebs 372 en.
pdf.
Gonzalez-Bailon, S., 2009. Opening the black box of link formation: social factors
underlying the structure of the web. Soc. Netw. 31 (4), 271–280.
Goodreau, S.M., Handcock, M.S., Hunter, D.R., Butts, C.T., Morris, M., 2008. A statnet
tutorial. J. Stat. Softw. 24 (9), 1–26.
Gould, R.V., Fernandez, R.M., 1989. Structures of mediation: a formal approach to
brokerage in transaction networks. Sociol. Methodol. 19 (1989), 89–126.
Grundmann, R., Scott, M., 2014. Disputed climate science in the media: do
countries matter? Public Underst. Sci. 23 (2), 220–235.
Guerra, P.H.C., Meira Jr, W., Cardie, C., Kleinberg, R., 2013. A measure of
polarization on social media networks based on community boundaries. Paper
Presented at the ICWSM.
Habermas, J., 1996. Between Facts and Norms: Contributions to a Discourse Theory
of Law and Democracy. MIT Press, Cambridge, MA.
Handcock, M.S., Hunter, D.R., Goodreau, S.M., Krivitsky, P., Morris, M., 2016. Ergm:
Fit, Simulate and Diagnose Exponential-Family Models for Networks.
11
Hargittai, E., Gallo, J., Kane, M., 2008. Cross-ideological discussions among
conservative and liberal bloggers. Public Choice 134 (1–2), 67–86.
Häussler, T., Adam, S., Schmid-Petri, H., Reber, U., 2017. How political conflict
shapes online spaces: A comparision of climate change hyperlink networks in
the United States and Germany. Int. J. Commun. 11, 3096–3117.
Hindman, M., 2008. The Myth of Digital Democracy. Princeton University Press,
Princeton.
Hunter, D.R., Handcock, M.S., Butts, C.T., Goodreau, S.M., Morris, M., 2008. ergm: a
package to fit, simulate and diagnose exponential-family models for networks.
J. Stat. Softw. 24 (3), 1–29.
Jost, G.F., Jacob, K., 2004. The climate change policy network in Germany. Eur.
Environ. 14 (1), 1–15.
Keck, M.E., Sikkink, K., 1999. Transnational advocacy networks in international and
regional politics. Int. Soc. Sci. J. 51 (159), 89–101.
Kriesi, H., 2004a. Political context and opportunity. In: Snow, D.A., Soule, S.A.,
Kriesi, H. (Eds.), The Blackwell Companion to Social Movements. Blackwell,
Oxford, pp. 67–90.
Kriesi, H., 2004b. Strategic political communication. Mobilizing public opinion in
‘audience democracies. In: Esser, F., Pfetsch, B. (Eds.), Comparing Political
Communication: Theories, Cases, and Challenges. Cambridge University Press,
Cambridge, pp. 184–212.
Krippendorff, K., 2012. Content Analysis: An Introduction to Its Methodology. Sage
Publications, London.
Li, W., Bradshaw, A.E., Clary, C.B., Cranmer, S.J., 2017. A three-degree horizon of
peace in the military alliance network. Sci. Adv. 3 (3), 1–10.
Lusher, D., Koskinen, J., Robins, G., 2012. Exponential Random Graph Models for
Social Networks: Theory, Methods, and Applications. Cambridge University
Press, Cambridge.
McPherson, M., Smith-Lovin, L., Cook, J.M., 2001. Birds of a feather: homophily in
social networks. Ann. Rev. Sociol., 415–444.
Morris, M., Handcock, M.S., Hunter, D.R., 2008. Specification of exponential-family
random graph models: terms and computational aspects. J. Stat. Softw. 24 (4),
1–24.
Myers, D.G., Lamm, H., 1976. The group polarization phenomenon. Psychol. Bull. 83
(4), 602–627.
Newman, M.E., Girvan, M., 2004. Finding and evaluating community structure in
networks. Phys. Rev. E 69 (2), 1–15.
Petricek, V., Escher, T., Cox, I.J., Margetts, H., 2006. The web structure of
e-government-developing a methodology for quantitative evaluation. Paper
Presented at the Proceedings of the 15th International Conference on World
Wide Web.
Pons, P., Latapy, M., 2005. Computing communities in large networks using
random walks. Paper Presented at the International Symposium on Computer
and Information Sciences.
Rogers, R., 2012. Mapping and the politics of web space Theory. Cult. Soc. 29 (4–5),
193–219.
Rogers, R., 2010. Mapping public web space with the Issuecrawler. In: Reber, B.,
Brossaud, C. (Eds.), Digital Cognitive Technologies: Epistemology and the
Knowledge Economy. John Wiley & Sons, Inc, Hoboken, NJ, pp. 89–99.
Stoner, J.A., 1961. A Comparison of Individual and Group Decision Involving Risk.
The MIT, Cambridge, MA (Master thesis).
Stroud, N.J., 2008. Media use and political predispositions: revisiting the concept of
selective exposure. Polit. Behav. 30 (3), 341–366.
Sunstein, C.R., 2009. Republic.com 2.0. Princeton University Press, Princeton.
Tremayne, M., 2005. News websites as gated cybercommunities. Convergence 11
(3), 28–39.
Weidner, H., Mez, L., 2008. German climate change policy a success story with
some flaws. J. Environ. Dev. 17 (4), 356–378.
Williams, H.T., McMurray, J.R., Kurz, T., Lambert, F.H., 2015. Network analysis
reveals open forums and echo chambers in social media discussions of climate
change. Glob. Environ. Change 32, 126–138.
Woo-young, C., Park, H.W., 2012. The network structure of the Korean
blogosphere. J. Comput.-Mediated Commun. 17 (2), 216–230.
Zaller, J., 1992. The Nature and Origins of Mass Opinion. Cambridge University
Press, Cambridge.
Please cite this article in press as: Häussler, T., Heating up the debate? Measuring fragmentation and polarisation in a German climate
change hyperlink network. Soc. Netw. (2017), https://doi.org/10.1016/j.socnet.2017.10.002
Документ
Категория
Без категории
Просмотров
0
Размер файла
1 463 Кб
Теги
2017, 002, socnet
1/--страниц
Пожаловаться на содержимое документа