Content uploaded by Max Frischknecht
Author content
All content in this area was uploaded by Max Frischknecht on Feb 15, 2022
Content may be subject to copyright.
Max Frischknecht
Master Thesis
HS2019/2020
January 9th, 2020
HKB BFH
Master Design
Design Research
Mentoring Ulrike Felsing,
Katherine Hepworth
Predicting Voters:
The Signicant Role of Personal Data
for Political Communication
in Switzerland’s Social Networks
1. Abstract
2. Research Plan
2.1 State of the Art
2.2 Current State of Our Research
2.3 Detailed Research Plan
2.3.1 Assumptions & Hypotheses
2.3.2 Research Question
2.3.3 Concrete Goals
2.3.4. Overview of Methods
Literature & Media Review
Expert Interviews
Web Content Analysis
Data Visualization
2.3.5 Work Packages (WP) & Milestones (M)
2.3.6 Timetable
2.3.7 Research Team
2.3.8 Risks
2.3.9 Budget
2.4 Relevance & Impact
2.5 Next Steps
3. Documentation
4. Declaration
5. Bibliography
2/29
Predicting Voters
1. Abtract
Social networks like Facebook, Twitter and others are becoming increasingly
important and enable novel approaches for political communication.
Simultaneously, the misuse of personal data is of rising concern for many
policymakers worldwide. Firstly, computational automation enables the
easy and quick dissemination of opinions, while false social media proles
create the impression of fake publicity. Secondly, personal data collected
from social networks offer political actors the possibility to predict the behavior
of their potential voters. Methods of psychology are used for target group
segmentation and are the basis for persuasive political adverts (Micro-Target-
ing) like Cambridge Analytica used during the US elections 2016. These
developments prompted universities around the globe to launch research
projects. However, studies that examine the signicance of personal data
for political communication in Switzerland are rare to nd. Although the
elections in October 2019 showed clearly that swiss parties increasingly use
personal data for their campaigns. There is little research that sheds light
on the involved actors, for example, the agencies and their relationships with
the parties.
Therefore, the project addresses the following research questions:How
do Switzerlands parties collect and utilize personal data from potential
voters to predict their attitudes, motivations, and behaviors? How does this
inuence their political communication in terms of design and message?
With whom do they collaborate in this process?
The project produces knowledge about the process of using personal
data for political communication on the example of Swiss ballot meetings
and analyses the involved actors and their relations. It proposes a combination
of different research methods to examine theoutcome (e.g. the advert on
Facebook)as well asthedecision-making process behind the advert. A web
content analysis creates a database of political adverts placed on social
networks by Swiss parties. A series of expert interviews illuminate the intentions
of the actors (e.g. parties, agencies) who created these ads. The collected
data is visualized in three thematic maps that depict the actors, their relations
and the political communication in the networks. The project is conducted
by an interdisciplinary research team from the elds of political science, data
science, and design research. The ndings will be disseminated in lec-
tures, peer-reviewed journals, and through an online open-source documenta-
tion. The produced knowledge is valuable for state institutions during
election observation or social networks preventing digital propaganda.
1. Abstract
3/29
Predicting Voters
2.1. State of the Art
Introduction
When compared to other countries, such as the United States, England or
Germany, Switzerland has been the subject of few studies that examine
the signicance of personal data for political communication in social networks.
In the following, the state of research on political communication and
manipulation in social networks will be introduced. This introduction consists
of two stages. The rst stage describes the global and local context and
concrete methods of political communication and manipulation. In the second
stage, the focus is directed onto the process of research itself and research
methods are discussed concerning their credibility. This involves
discussing data visualization and its importance for working with big data and
the contradictory role of using social networks as data sources.
Before introducing the state of the art, a few fundamental terms and concepts
should be briey claried to avoid misconceptions.
• Propaganda: The project understands Propaganda as „persuasive mass
communication that lters and frames the issues of the day in a way
that strongly favours particular interests; usually those of a government or
corporation“(Chandler & Munday 2011). This includes „the intentional
manipulation of public opinion through lies, half-truths, and the selective
re-telling of history“(Ibid.).
• Disinformation: is seen as a practice „involving the dissemination of false
information with the deliberate intent to deceive or mislead“(Chandler &
Munday 2011). Usually trough mass media.
• Mass Media: 1) the technological means of spreading messages and
information to a large, widely spread audience, 2) large-scale institutions
that produce and disseminate these messages. Following these
denitions, social networks like Facebook and Twitter count as mass
media and using them with the intent to inuence public opinion in
favor of particular interests can be seen as propaganda.
• Personal Data: Denes as „any information that relates to an identied or
identiable living individual1“. An individual is ‘identiable’ if it is
distinguishable from other persons. This also includes pieces of
information which, when brought together, can lead to the identication
of a person. Concerning behavior in a social network, every activity
(sharing, commenting, liking, posting, scrolling), can be measured and
put into relation with the individual, therefore, becoming personal data.
As a result, when politically contextualized, this data becomes
signicant for political communication.
• Anonymised Data: Contrary to personal data, this is data where it is
impossible to identify individuals. The anonymization must be proven
to be irreversible to be anonymous. Meaning it must not be possible to
identify the persons again by reengineering the anonymization process.
• Legal situation: It is difcult to make general statements about the
legal situation. One reason for this is that the respective platforms (e.g.
Facebook) are usually not liable in the respective country (e.g.
Switzerland). The users of the platforms are usually not aware that they
agree to the relatively lax use of their data by accepting the terms
and conditions. For example, Cambridge Analytica illegally collected
data while the Swiss parties remained within the framework of the
local laws. However, it is also the case that software used by some Swiss
parties, like Nationbuilder, is again prohibited in France (Püntener 2019).
1 European Commission - Eu-
ropean Commission. “What Is Personal
Data?” Text. Accessed December 4,
2019. https://ec.europa.eu/info/law/
law-topic/data-protection/reform/
what-personal-data_en.
2.1 State of the Art
4/29
Predicting Voters
Part 1: Political Communication in Social Networks
The Global Context
The use of automatization, algorithms and big data analysis with the intent to
inuence the political opinion is dened by the Oxford Internet Institute
(OII) as „Computational Propaganda“ (Bradshaw & Howard 2018). Political and
private actors who carry out such propaganda campaigns are termed
„Cyber Troops“ (Ibid.). The number of affected countries constantly rises and
has reached 70 by 2019 (Bradshaw & Howard 2019). Global spending on
such campaigns is more than half a billion dollars for the period between 2010
and 2018 (Bradshaw & Howard 2018). The rst evidence of the use of
Computational Propaganda dates back to 2010 (Bradshaw & Howard 2017),
however, conceptually similar political strategies, like the defamation of
oppositional candidates, are much older (Grassegger 2019). Today the
availability of large amounts of personal data from potential voters enables
political actors to create potentially more persuasive and purposeful messages
as we will see throughout this chapter. Broader public awareness gained the
phenomenon trough a worldwide scandal produced by the consulting agency
Cambridge Analytica. The rm harvested personal data from Facebook to
create psychologically-manipulative digital advertising for the US presidential
elections in 2016 and the Brexit vote (Davies 2015; Krogerus & Grassegger
2016). This incident led to some improvements like the creation of Facebook’s
Ads Library, where it’s possible to look up political advertisements.
However, the technological preconditions develop daily and call out for an
intense, ongoing commitment. Therefore, institutions around the globe
launched research projects to investigate the matter on an ongoing basis. The
OII published a series of global reports (Bradshaw & Howard 2017, 2018,
2019) and a set of country-specic reports on Mexico (Glowacki et al. 2018),
Sweden (Hedman et al. 2018), the United States (Howard et al. 2017),
Brazil (Machado et al. 2018), Germany (Neudert et al. 2017), Ukraine (Zhdanova
& Orlova 2019), and many others. The British Information Commissioner's
Ofce (ICO) published reports on the Brexit case (ICO 2018a, 2018b) and New
York’s University initialized the Social Media and Political Participation Lab.
The Local Context
Despite the described range of research by global institutions, examinations
with a focus on the Swiss context are seldom. One exception is the by
the Swiss Radio and Television (SRF) published podcast Hotspot2 and short
recurring reports in the news show 10vor103 during the votes in October
2019. These reports made clear, that the global developments don’t stop at
Switzerland’s border. The lates votes in October 2019 showed unequivocally,
that personal data becomes a driving force in Switzerland's political
campaigning (Püntener 2019). Almost every major party used personal data
on possible target groups to inform their communication strategy. Be it to
decide on which door to knock, which number to call, or, in the case of the FDP,
which image-text combination to show to which Facebook user, personal
data served as a basis. The reports by the SRF and own research (cf. 2.2)
showed, that working with data is a complex procedure. It requires various tech-
nical specializations to collect and utilize personal data from potential
voters. Consequently, this process also involves different actors, including
parties, agencies, data brokers, creators of proling models and the plat-
forms (e.g. Facebook). However, research examining these actors and their
roles in regards to their inuence on political communication is yet rare
to nd. In line with the global trend, spending on digital campaigns is rising
among Swiss parties. The mother party of the CVP, for example, invested
260’000 CHF in a digital campaign (Ruch 2019). The success of the Grüne in
October 2019, who resigned from working with data because of privacy
concerns, also shows that data-based campaigning is not automatically a
recipe for success. Experts assume that the trend will continue and the parties
will intensify their engagement with data, assumingly becoming more
effective in using it (Püntener 2019). Since these developments happened very
recently, it does not surprise, that systematic research on Switzerlands situation
2 https://www.srf.ch/play/
suche?query=hotspot%20datenspur,
accessed January 9th
3 https://www.srf.ch/sen-
dungen/10vor10/wahlkampf-digital,
accessed January 9th
2.1 State of the Art
5/29
Predicting Voters
is still seldom. The high likeliness of an ongoing, intensied engagement of
the parties with personal data calls out for an investigation that enables a broad
public discussion around personal data becoming a political commodity.
Now, we start to grasp the global, local, and nancial dimensions of the
topic. But we didn’t look at concrete methods of political communication
that prot from personal data and computation. The following sections will
introduce some of these methods, however, it is beyond the means of
this proposal to create a concluded description of all methods. Rather should
two prominent ones, Micro-Targeting, and Social Bots, be described in terms of
their state of the art and their role in the Swiss context.
FakeProlesandSocialBots
One practice of computer-aided propaganda that gained much attention is the
creation of false social media proles, commonly called Bots or Trolls.
According to Facebook, such accounts make up 3-4% of all users. Between
January and March 2018, Facebook deactivated 538 million of fake accounts.
And according to an analysis by cybersecurity rm Imperva, bots accounted for
about half of global internet trafc in 2015 (Zweifman 2015).
Of course, not all bots are politically motivated. False or articial social
media proles, also in the political context, are summarized under the
collective term Social Bots. When organized in large quantities, these proles
can give the impression that many, supposedly real citizens, have a certain
political opinion. In a political context, they are dened as: „the algorithms that
operate over social media, written to learn from and mimic real people so as
to manipulate public opinion across a diverse range of social media and device
networks.“ (Woolley & Howard 2016, 4885). Social bots can be divided into
three basic types:
1. Bots:Purely articial proles. All their activities are evoked through
computer programs. This includes, for example, the retweeting or sharing
of articles from certain politically inuenced media portals or the
distribution of fake and junk news. Bot activities have been detected in 38
countries. (Bradshaw & Howard 2018).
2. Trolls: Proles operated by humans. They take part in online discussions
trying to inuence public opinion. A popular example is the „50 Cent
Party“ which operates on behalf of the Chinese state (King et al. 2017).
Trolls are often used to suppress or defame certain opinions through hate
speech. Several trolls working together are called a „troll factory“.
Such movements emerge in a decentralized way over chatrooms or
forums. Investigations into the GamerGate affair have shown that there
are detailed tutorials available that explain to users how to create
false Twitter proles and effectively combat opposite opinions (Trice &
Potts 2018).
3. Cyborgs: A combination of articial and human-based activities. Because
of their human component, these bots are especially hard to detect.
The OII found evidence of cyborg activities in 9 countries (Bradshaw &
Howard 2018).
Depending on the case and the research, further or other bot categorizations
are made, including ImpactBots,Ampliers,Dampeners,Complainers,
Trackers, or ServiceBots (Zhdanova & Orlova 2019; Dubois & McKelvey 2019).
The detection of bots, meaning the distinction from a prole operated by a
real human being, is not always easy to achieve. A fundamental feature is that
bots do not produce genuine content of their own, but, for example,
simply distribute the content of a news page. The choice of the news page,
however, indicates the political motivation behind the bot (Sanovich et al. 2018)
(cf. Fig. 1).
2.1 State of the Art
6/29
Predicting Voters
A series of private and public institutions specialized in the observation and
recognition of bots, for example, Botswatch4 or the Observatory on Social
Media5. It must be noted that bots are not fundamentally something negative.
For example, they are actively used by journalists to collect data or monitor
government websites for news (Dubois & McKelvey 2019). I also used a bot to
collect tweets from the Swiss parties during the votes in October 2019
(cf 2.2). Till now there are no indications of political bots being active in Switzer-
land. However, there has not yet been much systematic research on the
usage of political bots in Switzerland. It is to note that, even though bots have
received a lot of attention, their in uence is controversial and many experts
believe that bots are less in uential than often believed (Reuter 2019).
Additionally, since personal data doesn’t play a foreseeable role in the usage of
bots, the project proposes to turn the focus towards a much more subversive
and persuasive way of communication: Micro-Targeting.
Micro-Targeting
Micro-Targeting is an advertising technique that is increasingly used in political
communication. It aims to de ne highly speci c target groups based on
unique data about that group. In contrast to conventional target group speci c
advertising, this data goes far beyond socio-demographic characteristics.
Not age, income or gender are of primary interest, but attitudes, motivations,
and behavior. Classic advertisement tries to reach as many people as
possible (e.g. women over 30, with children and average income from the Bern
region). Micro-Targeting aims for the opposite, reaching the smallest de n-
able groups (e.g. women whose parents are divorced, like punk music and eat
vegetarian food). Based on this information the (micro-)targeted advertise-
ment, which can be of various media like a mailing, online advert or phone call,
is crafted in a way that the data suggests is likely to resonate.
Consequently, a fundamental part of Micro-Targeting is the linking of
large amounts of data from potential targets with methods of group
segmentation and pro ling. For this purpose, the political actors use methods
from psychology as the case of Cambridge Analytica and the Swiss votes
showed (Krogerus & Grassegger 2016; Püntener 2019). These so-calledpro ling
modelshelp to undertake the data based group segmentation and to make
predictions about how likely a target is to respond towards a possible advert. It
is assumed that the more personal such data is, the higher the accuracy of
segmentation and prediction is. Cambridge Analytica used the Five-Factor-
Model (FFM) for this purpose. The FFM is well situated in psychology since the
1980s and de nes personality traits along the ve basic dimensions of
openness, conscientiousness, extraversion, agreeableness, and neuroticism
(McCrae & Costa 1997). It’s important to understand that, today, social
media provides psychology with unprecedented amounts of personal, freely
available, data. Because of this, researchers achieved to predict personal
attributes like gender, sexuality or political orientation based on Facebook-Likes
with an accuracy of 80% to 90% (Kosinski et al. 2013). They state that they
can predict a person better than the person's partner based on a data set of 200
likes (Popov 2017). Cambridge Analytica used this exact method to create
Fig. 1 A bot classi ed as an Ampli er
because it only retweets content from
news sources on a high frequency
(about 64 tweets per day) (Zhdanova &
Orlova 2019)
4 https://botswatch.de
5 https://botometer.iuni.
iu.edu/#!/
2.1 State of the Art
7/29
Predicting Voters
Micro-Targeting ads addressing potential voters based on the predictions made
from that voter's Facebook-Likes (Krogerus & Grassegger 2016). The agency
illegally collected the data on Facebook through a fake play game including a
hidden personality test. The Swiss data broker Schober AG, who sold data to
the CVP and FDP, uses a similar model to the FFM called „Sinus Milieus“ for its
analysis (Püntener 2019). The model's typology divides target groups based
on „values and views of life“ and „social situation“. A brief analysis of political
advertising on Facebook that I undertook during my studies has shown that, for
example, the FDP places ads that suggest the use of proling methods.
These ads work with a dened set of images and different assigned texts
to alter the message. For example, one of the advertisements argues from
an ecological perspective and the other from a more economic perspective (cf.
Fig. 2–3 and 2.2The Network II). I assume that these different meanings target
potential voter groups segmented based on personal data.We recognize,
that using personal data for Micro-Targeting requires various specializa-
tions like data collection and analysis and brings different actors into
play: the creator of the proling model (e.g. the Sinus Institute), the data
broker (e.g. Schober AG) and the agency designing the adverts.
We can conclude that computational methods for political communication and
manipulation are on the rise worldwide and in Switzerland. While social bots are
a large topic, the use and misuse of personal data is of great signicance.
Research that examines how Swiss parties and other involved actors (plat-
forms, data brokers, creators of proling models and agencies)collect and
utilizepersonal data to predictattitudes, motivations and behaviorfrom
potential voters is yet rare to nd. Not much is known about thedecision-making
processthat leads to the design of different adverts who address users
individually. However, this is by no means an easy research undertaking,
therefore we will look at some research methodologies that analyze
communication in social networks regarding their credibility in the next section.
Part 2: The Process of Research
Data Visualisation: A Necessary Research Tool
When examining political communication in social networks, researchers end
up with large amounts of data, for example, Tweets in the millions. As a
result, many researchers use data visualization as a tool to depict results and
uncover patterns in data. As Katherine Hepworth, a specialist in the ethics
of visualization notes: „Working with big data shifts the role of data visualization
in research. It moves from an optional research output to a necessity for
data exploration.“ (Hepworth 2017, 12). While it is possible to manually examine
a few hundred data points, it becomes a „physiological impossibility“ if the
set consists of millions (Ibid. 12). Data visualizations have proven to be helpful in
a wide range of research cases (Tufte 1990). They even have a quite long history
Fig 2–3 Political adverts by the FDP
placed on Facebook during the votes in
October 2019.
2.1 State of the Art
8/29
Predicting Voters
of depicting political power structures and communication patterns (Pfeffer
2017). And they are also used regularly for researching political communication
and manipulation (cf. Glowacki et al. 2018; Machado et al. 2018).
However, truthful data visualizations are by no means a self-evident fact.
It is rather the case that creating data visualizations must be seen as a complex
procedure of engineering, storytelling, and argumentation (Tufte 2001).
This process ranges from data collection, revision, analysis, interpretation to
nal visualization and involves many decisions made by the researcher. These
decisions ultimately inuence the nal argument made by the visualization
(Hepworth & Church 2018). Johanna Drucker highlights that „by rendering
statistical information into graphical form, a simplicity and legibility is created
that hides every aspect of the original interpretative framework on which
the statistical data were constructed“ (Drucker 2011, sec. 8). The effectiveness
of visualization, as Drucker criticizes, has caused humanities researchers to
lose the necessary criticism for rigorous scholarly research: „The sheer power of
the graphical display of ‚information visualization‘ […] seems to have produced
a momentary blindness among practitioners who would never tolerate such
literal assumptions in textual work“ (Ibid., sec. 5.). What Drucker describes as an
„interpretative framework“ refers to the many steps, including data collection,
revision, analysis, interpretation, and visualization, to create a chart or map.
All the small decisions along the way sum up and formulate the argument of the
visualization. Therefore a visualization can’t be seen as a simple representa-
tion of pre-existing facts as we usually assume. Striving for high standards when
creating data visualizations seems like an inescapable necessity, even an
ethical duty as some argue (Cairo 2019). This, as I would like to add, holds espe-
cially true when visualizing political data in the current climate of distrust,
fake news and increasing disbelief towards scientic ndings. As we will see
soon, the described problems can also be detected in data visualizations
depicting results from research on political manipulation in social networks. But
before we examine such a visualization of political communication, let’s
quickly look at some attempts to achieve the described standards and how
they could help our research.
One way is to consider the latest guidelines from the eld like the by
Hepworth (2019) developed Ethical Data Visualization Workow. Although it is
not specically designed for political content, it can serve as a starting point.
Other approaches aim at clearly understanding and contextualizing the data
source. This especially includes acknowledging and depicting the fact that
data sets are rarely 100% certain. This can have different reasons: missing
values, sources of low credibility, errors introduced during storage or
computational processes, deviations or lack of precision (Griethe & Schumann
2005). Different concepts and typologies try to detect, categorize, and measure
uncertainty in data (MacEachren et al. 2005; Thomson et al. 2005). However,
these approaches are mainly developed for specialized domains like
cartography. They don’t emphasize on the topics of social networks in a
political context. Nevertheless, these theories help us to detect one problem
that visualizations of political communication and manipulation usually have:
the uncertainty
of the data source. Let us look at this data source and an visualization example
in more detail to concretize the mentioned problems.
APIs: Controversial Datasources
To make a data visualization you need data. And when visualizing political
communication in social networks most studies collect the data from the
network itself. This collection is made possible by accessing the network's
Application Programming Interface (API). APIs enable us to communicate with
external software (e.g. Twitter) trough code to send and receive data. For
example, you conduct a search for the hashtag „wahlen2019“ and the Twitter
API will return Tweets containing it. You can also receive data about a party's,
candidate's or user's prole, and information about the content they share. This
data is analyzed by researchers and, for example, visualized in the form of a
map of the political afliation of Twitter proles (Fig. 4).
2.1 State of the Art
9/29
Predicting Voters
To draw the above Fig. 4 Machado et al. used the Twitter API as a data source
and the commercial visualization software Graphika6. However, the
representative character of APIs as data sources is controversial. Machado et al.
note in their paper: „The platform’s precise sampling method is not disclosed,
however, Twitter reports that data available through the Streaming API is, at
most, 1% of the overall global public trafc on Twitter at any given time“ (2018,
3). This means that the API only offers a fraction (1%) of all the possible data
from Twitter. And what the decision-making process is, of a data point
becoming part of this 1% or not, remains unknown. It is, therefore, reasonable
to ask, how representative is 1% in regards to the political communication on
Twitter? Justiably there are critical voices from the research community: „A
data set made up of any Tweets […] might contain important public discourse
about relations […] (but), it is impossible to determine if those Tweets re-
present all such discourse, what universe of discourse they came from, or the
nature of that universe. […] The representativeness of non-random samples,
drawn from an unknowable universe, is pure conjecture“ (Lacy et al. 2015, 3).
The composition of 1% of the data remains unknown. Often referred to as
„Black Box“, few people understand what is happening inside the algorithms
of big data computation and this also holds true for APIs. The reason for this is
that the platforms don't disclose how their algorithms work. Nevertheless,
an API is one of the few ways to access a social network. Therefore, API’s as
data sources can’t be avoided completely.
I would like to argue, that we need to shift the focus towards a respons-
ible contextualization of the collected data. This includesshowing these
limitations in visualizations. If you look at Fig. 4 without having read this
section you would most likely assume that it represents 100%. All the
information you need to comprehensibly „read“ the visualization isn’t depicted
but hidden in the corresponding text. To be able to pay attention to such details
in visualizations you need experts from the eld of design.
It can be summarised, that data visualizations are a necessity when working
with data sets in the millions. The involved steps of data collection, revision,
analysis and interpretation shape the argument of the visualization. Data
collection on social networks is mostly undertaken through the networks API.
However, this API is also a „Black Box“ since it’s unclear which data can be
collected and which not.This uncertainty is usually not visible in the
visualization.I argue, that in a context of distrust, fake news and increasing
disbelief towards scientic ndings,researchers are obliged to show these
limitations in their visualizations, and not „hide” them like it’s done in
Fig. 4. Otherwise, it's an easy undertaking for critics to doubt research results.
To counteract these limitations, research that examines political
communication and manipulation in social networks should not focus solely on
APIs, but rather expand its sources through conducting expert interviews
with the involved actors (parties, agencies, data brokers, etc.). This approach
makes it possible to look at the outcome (e.g. the Facebook advert) but also
learn about thedecision-making process behind the advert.
Fig. 4 Twitter proles afliated with
possible candidates by color and
position based on the content these
proles share and like. Each node (dot)
represents a prole while its size deter-
mines the amount of followers (reach)
(Machado et al 2018)
6 https://www.graphika.com/
2.1 State of the Art
10/29
Predicting Voters
2.2 Current State of Our Research
At the current stage, this section summarises research conducted during my
MA studies in the eld of design. At a later point, this section also includes the
state of research from the other members of the research team (cf 2.3.6).
Design Research
During my studies at the University of Arts Bern I examined the political
dimension of digitization from four perspectives: 1) the current state of research
on Computational Propaganda, Micro-Targeting, and Bots, 2) critical studies
on ethical data visualizations, 3) debates on the scientic soundness of digital
data sources, for example, APIs, 4) the necessary technological skills to collect,
process and visualize digital data, and 5) artistic inquiries from the eld of
political Media Arts. This acquired knowledge could already, and will further be
disseminated trough lectures and workshops:
• "P5js for Beginners", Workshop at FHNW Basel, with
Mark Iandovka, February 2019
• "0/1 – Black/White – On/Off – Dead/Alive", Workshop at
ETH Zürich, March 2019
• "Copy It Right! On sharing, caring, data and network politics",
Workshop, Lecture at Masterstudio Design, FHNW Basel, with
Yann Martins, March 2019
• "How to visualize computational propaganda", Lecture at
Interaction Design ZHdK Zürich, April 2019
• "Processing", Workshop at Hyperwerk FHNW Basel, with
Mark Iandovka, April 2019
• "Fundamentals in digital design practice", Workshop at
BA Visual Communication FHNW Basel, February 2020
ScriptedLoopholesSeries
The ve parts of this ongoing series were developed throughout the MA studies
and provided a space for technological experimentation. They helped me
to grasp the possibilities and limitations of network-related data and to deepen
my programming and design skills. The extension of this engagement can be
summarised by three main aspects7:
1. How and where to collect digital data through the usage of Application
Programming Interfaces (APIs). This resulted in experiments with six
different APIs, namely the Reuters News API, Twitter API, Wikipedia API,
and the by Google developed Geocoding, Maps and Cloud Natural
Language API.
2. Learning the necessary technologies to collect, store, revise, analyze
and visualize data. This included learning NodeJs, VueJs, D3Js, Git,
Heroku and becoming familiar with standard data formats (JSON, CSV).
3. Learning about the limitations of these technologies and more
sophisticated data science approaches (Python, Tensor Flow).
Following, the ve parts of the series and their relevance for the proposed
project should be briey introduced.
TheSearchEnginewas developed as an assisting tool for my literature and
media review. It browses the Reuters database for news articles related to
computational propaganda and creates a daily updated news feed. The tool
provided rst knowledge about how to access an API and how to process
and visualize collected data. Through the technical reconstruction of a search
engine, it became clear, that search engines like Google can’t be regarded
as neutral gateways (L'Ecuyer et al. 2018). The search results that Google
delivers depends on its construction process. The experiment is a starting point
for creating alternative channels of information collected during the proposed
literature and media review (cf. 2.3.4). Most leading news and academic outlets
(e.g. JSTOR) maintain an API to bypass default search engines.
7 More detailed informations
can be found in the documentation
under „Technology Excursion“
2.2 Current Stage of
our Research
11/29
Predicting Voters
The Advertiser analyses the Facebook Ads Interests, a list that Facebook
creates for every prole. This list is partly based on direct user behavior (liking,
commenting), and partly predicted by a „Black Box“ algorithm. The tool
calculates how many entries Facebook created autonomously by comparing
the list to my likes:
• Advertisement interests in total: 218
• Liked by me: 32
• Dened by Facebook: 186
Facebook’s prediction of my interests was sometimes true but mostly com-
pletely wrong. It supported the hypothesis that algorithms are good at
collecting but bad at contextualizing. Nevertheless, I still believe it’s possible to
make assumptions about my attitude, motivations, and behavior from this
list. A relatively large amount of terms in the list where unknown to me,
therefore, the tool includes an automated Wikipedia summary for every term.
The tool helped to deepen the knowledge about social network-related data
and how to analyze it.
TheMessengeris inspired by the uncovering of Facebook selling private chat
records to companies like Netix and Amazon (Newton 2018). Such chat
records can be of great length and I assume that the companies use algorithms
to extract useful information. I was curious how much information an
algorithm could derive from a chat between a friend of mine and me. Inspired
by the discovery of my ad interests inThe Advertiser, I chose to run an entity
analysis on my chat with a Google-developed machine learning algorithm. The
algorithm identies dates, persons, contact information, organizations,
locations, events, products, and media types in unstructured text. Surprisingly
the produced results consisted of almost 90% noise (wrong predictions).
Nevertheless, it was possible to assume the personal address, profession, and
country of origin of my friend from the results. The tool helped to create
knowledge on machine learning-based APIs and text analysis. It could be
potentially helpful for data analysis of expert interview transcripts (cf. 2.3.4).
The Network is an ongoing project to visualize political activity on Twitter during
the Swiss votes. The tool uses the Twitter API and supports the thesis that
APIs only provide a small insight into a network (cf. 2.1 APIs). Between the 17th
and 20th of October, I collected around 12’000 Tweets with the help of a
data collection script. Simultaneously this provided me with the necessary skills
to program a Twitter-Bot. The search query consisted of political proles from
parties and candidates and political hashtags like „#jetztFDPwählen“ or
„#darumstarkeSP“. The collected Tweets were pruned and geographically map-
ped across Switzerland. The project has shown that more complex under-
takings usually require the use of different APIs to collect, analyze and visualize
data. For example, was the Twitter API used to collect data, while two APIs
by Google helped to dene the Tweet's geographic origin and placing it on the
map of Switzerland. A more detailed description of the process can be found
in the documentation (cf. SL4 The Network Logs 1-6).
TheNetworkIIis an ongoing investigation of the Facebook Ads Library.
Research by the SRF has shown, that the FDP obtained data from Schober AG
via the opinion research institute GSF Bern (Püntener 2019). In the advertising
library of Facebook, various Swiss election advertising can be found. A
series of adverts placed by the FDP show several different design alternatives
making usage of versatile image-text combinations (Fig. 5-8). I assume that
the different arguments of the advertisements aimed at different (micro-) target
groups. The by Schober AG used „Sinus Model“ (cf. 2.1Micro-Targeting)
could have served for the segmentation of the groups. These ndings serve as
a starting point for a preliminary study that I will submit to the BFH Call for
Proposals 2021 (cf 2.5).
2.2 Current Stage of
our Research
12/29
Predicting Voters
Relevance for the research project
The engagement with the Scripted Loopholes series supports the hypothesis
that the collection, analysis, and visualization of data is a complex proce-
dure calling for an interdisciplinary team of specialists. Supporting the thoughts
of Drucker (2011), Hepworth & Church (2018), collecting data and creating
truthful data visualizations is by no means a self-evident process. The more com-
plex an investigation becomes, the more technologies are required for
processing. Each of these technologies, in turn, has the potential to have an
impact on the nature of the data and the argument of the visualization.
The rst four parts of the series can be visited together with the documentation
under: https://males.maxfrischknecht.ch
Fig. 5–8 Political adverts by the FDP
placed on Facebook during the votes in
October 2019.
2.2 Current Stage of
our Research
13/29
Predicting Voters
2.3 Detailed Research Plan
2.3.1 Assumptions & Hypotheses
State of the art and the own research ndings lead me to the following
hypotheses: Personal data becomes signicant for the political communication
of the Swiss parties while the involved actors and their methods of collecting
and utilizing personal data are intransparent. This lack of transparency poses
the danger, that a switch from contemporary data-informed campaigning into
computational propaganda goes unnoticed by the public. The line between
legally using the available technological possibilities for political purposes and
the unethical, manipulative misuse of personal data is thin and needs to be
dened. The role of personal data differs very much based on the involved
actors and the specics of data collection and utilization, therefore, knowledge
applicable to the swiss context is limited. To improve transparency it is
necessary to acquire the following knowledge trough rigorous on-site research:
• If we want to understand the role of personal data for political campaigns
we can’t only look at the outcome (e.g. an advert on Facebook), we also
need to investigate the circumstances and the decision-making
processes that created this advert.
• Many actors with different roles are involved in this process and it's
necessary to identify them. At this stage, the following actors have been
dened:
• Users of social networks who spread personal data
• Platforms who collect, analyze and sell the data of their users
(e.g. Facebook)
• Data brokers who aggregate, analyze and sell data (e.g. Schober
AG)
• Creatorsofprolingmodels that help to undertake the data
based group segmentation (e.g. Sinus Institute)
• Communication agencies and advisors who conceptualize and
design the communication (e.g. Farner, Enigma)
• The party leadership that develops and implements the overall
strategy
• Individual candidates of the parties who develop independent
communication strategies for themselves
• These actors have different motivations and expectations for working
with data. While the creator of a proling model might be led by
scientic interest, the data broker could be guided by economic and the
party by political interests. This inuences how they process the data
because they expect different results from working with it.
• It’s important to understand how personal data enables the actors to
make predictions about a person’s attitude, motivation, and behavior.
• The ability to make such predictions changes the design process of the
adverts. Without this data the agencies have to guess what resonates
with the voters, now they can claim that they know beforehand.
• It’s important to understand the transformative processes and the different
characteristics of the data during the process. While it could be the case,
that personal data used to create a proling model (e.g. the Sinus Model) is
anonymized, therefore not personal anymore, the data used by a data
broker to feed the model, however, is personal.
• It's important to understand what characterizes the relationships these
actors have with each other, are they commercially, politically or
otherwise? For example, it would be critical if an institute with a public
mandate would share its ndings with a party.
• Understanding the intentions and the processes behind the outcome (e.g.
the Facebook advert) are much more valuable than just analyzing the
different outcomes itself. It leads to a more profound understanding of the
meaning and power of personal data for political communication by
looking at the decision-making process. While analyzing only the outcome
leaves the core of the matter blurry and in the worst-case contributes to
polarising opinions. An understanding of the background, therefore, is
also more likely to contribute to a differentiated debate.
2.3 Detailed Research
Plan
14/29
Predicting Voters
These assumptions are the basis for the Actors Map (Fig. 9). It shows the
presently identi ed actors, their relationships and the processes of collection,
analysis, interpretation and communication. This visualisation serves the project
as a starting point (cf 2.3.4 Data Visualization).
2.3.2 Research Question
HowdoSwitzerlandspartiescollectandutilizepersonaldatafrompotential
voterstopredicttheirattitudes,motivations,andbehaviors?Howdoesthis
influencetheirpoliticalcommunicationintermsofdesignandmessage?With
whomdotheycollaborateinthisprocess?
To answer the overall questions the project formulates a set of subquestions:
• What actors are involved in the process of, and how do they collect,
revise, analyze, and interpret personal data for political communication?
• What are the expectations and motivations of the actors to work with
personal data?
• How is the character of the data at the stages of collection, revision,
analysis, and interpretation, and how does it transform from one stage to
the other? (Is it always personal? When does it become personal?)
• What conclusions are drawn from personal data and how does it
in uence their political communication in terms of design and message?
• What constitutes the relationship the actors have with each other, are
these commercially, politically or otherwise?
2.3.3 Concrete Goals
The proposed project systematically explores the role of personal data for
political campaigning in Switzerland. It produces detailed and genuine know-
ledge about the processes of datacollection, revision, analysis, and
interpretationof the example of Swiss ballot meetings. Furthermore, the
project produces knowledge about how personal data enables the parties to
predictattitudes, motivations, and behaviors from potential votersand how
thisshapes the political communication in terms of design and
message(e.g. a Facebook advert). The project clari eswhich actors are
involvedin this process, what theirexpectations and motivationsare to work
with personal data in a political context andwhat characterizes the
Fig. 9 Actors Map
User
I
n
s
t
a
g
r
a
m
P
r
o
f
i
l
e
F
a
c
e
b
o
o
k
P
r
o
f
i
l
e
T
w
i
t
t
e
r
P
r
o
f
i
l
e
Plaform Individual
Data Collection
Advertisment
Interface with User
Target Groups
Data Collection
Tools (Social
Listening)
Listening)
Listening)
Data Broker
Academia
Agency
Political Party
Plaorm
Profiling Models
(FFM, Sinus)
Psychologically
Segtmented
Voter/Topic Groups
Data Collection Process
Advertisment Creation ProcessAdvertisment Interface Creation Process
Data Analysation &
User Segmentation Process
Political Party
L
a
r
g
e
s
e
t
o
f
P
e
r
s
o
n
a
l
D
a
t
a
&
R
e
l
e
v
a
n
t
T
o
p
i
c
s
A
B
D
C
E
F
B
Likes
Tweets
Images
Friends
Public Data
Personal Data
Relevant Topics
Micro Targeting Adverts
Personal Data
Relevant Topics
Psychology Institutes
(Sinus, Cambridge,
Stanford)
Advert that matches
your psychological
profile
Contacts
Images
Behaviour
Messages
The Cycle of Data
Collection & Utilization
I
n
s
P
r
2.3 Detailed Research
Plan
15/29
Predicting Voters
relationship between the different actors. In this way, the project contributes
to a concrete understanding of thepolitical signicance of personal data in
Switzerlandand places the Swiss situation in a global context. Besides, the
project develops new subject-specic knowledge onhow to classify APIs as
data sourcesand develops data visualizations that take the fact into account
thatdata is constructed and not the representation of a priori existing facts.
Thus it contributes to discourses in the elds of Political Science, Data Science,
Design Research and the Digital Humanities. The research team will
disseminate the ndings in lectures at international congresses, in
peer-reviewed journals, in the form of online open-source documentation and
the nal publication.
2.3.4 Overview of Methods
Literature & Media Review (WP1)
The literature and media review helps to develop theoretical foundations
and to rene hypotheses. It provides technological knowledge (facts and dates)
by reviewing literature and key documents. It addresses the following issues:
1. Understanding the global and local context of the role of personal data in
political communication. This involves reviewing academic and scholarly
studies8, government reports9, research conducted by civil society
organizations10 and investigative journalistic reports11. It helps the project
to assess the signicance of personal data and to understand the
methods of political communication that utilize personal data (e.g.
Micro-Targeting). It guarantees the last topicality of the research, and
allows the embedding of the ndings in a global context, thereby
contributing to the ongoing academic discussion in the eld.
2. Understanding the current state of possibilities to retrieve or buy personal
data in Switzerland and available offers for accompanying services
like analysis, utilization, communication, and design. Trough reviewing
service offerings and documents from platforms, data brokers,
creators of proling models and communication agencies12 the Actors-
Map (cf. page 14) will be rened. This deepens the hypotheses on the
involved actors, their roles, expectations, and motivations to work with
personal data.
The Actors Map and hypotheses developed upon the review lay the basis for
the expert interviews. It enables the project to put the focus of the following
expert interviews on process and interpretation knowledge and less on
technical knowledge, which, according to Bogner et al. (2014, 18) is consistent
with the methodological strengths of the method. However, review and
interviews are not understood as linear and consecutive, but rather as mutually
inuencing and iterative processes.
Expert Interviews (WP1 / WP2 / WP3)
22 expert interviews will be conducted with the following representatives:
• 3 different data brokers (e.g. Schober, Novalytica, Datahouse)
• 3 creators of proling models (e.g. the Sinus Institute, TBD, TBD)
• 3 different agencies (e.g. Farner, Enigma, Furrerhugi)
• 1 leading representative of each party (SVP, FDP, CVP, GLP, SP, Grüne)
• 1 candidate that works with data in his campaigns (SVP, FDP, CVP, GLP,
SP, Grüne)
• 1 representative of each platform (Facebook, Twitter)
The expert interviews provide knowledge about the decision-making process
behind political communication (e.g. a Facebook advert). It helps to under-
stand the role of the different actors and how they collect and utilize personal
data to predict attitudes, motivations, and behaviors from potential voters.
It also provides insight into the actor’s expectations and motivations to work
with personal data and their relationships with each other. These revelations
and roles will be visualized in the Actors Map.
8 e.g. Oxford Internet Institute,
Political Data Science University of Mu-
nich, Social Science Research Council
9 e.g. ICO
10 e.g. Algorithmwatch, Mozilla
Foundation, Digitale Gesellschaft,
Chaos Computer Club, Ranking Digital
Rights
11 e.g. Netzpolitk, Republik, De-
partment of Data Journalism at the SRF
where I already established a contact.
12 e.g. Novalityca, Datahouse,
Schober AG, GSF Bern, Nationbuilder,
Sinus Institute, Farner Consulting,
Enigma, Furrerhugi, Facebook/Twitter
for Business
2.3 Detailed Research
Plan
16/29
Predicting Voters
The expert interview is a method of qualitative social research whereby experts
are dened as persons who possess knowledge that is „particularly effective
in practice and thus becomes a guide for orientation and action for other
actors“ (Bogner et al. 2014, 14). Three different types of expert interviews will be
conducted to open up different levels of knowledge:
• Explorative Expert Interviews (WP1) serve as orientation in the eld and
for scrutinizing hypotheses.
• Systematizing Expert Interviews (WP2) try to collect the technological
and process knowledge comprehensively and analytically. Concretely
they will reveal how the actors collect and utilize personal data for
political communication to predict attitudes, motivations, and behaviors
from potential voters.
• Theorizing Expert Interviews (WP3) aim at the subjective dimension
(implicit decision, perceptual patterns, world views, etc.). They will reveal
the actor’s expectations and motivations to work with personal data and
the relationship the actors have with each other (commercially, politically
or otherwise).
Web Content Analysis (WP1 / WP2 / WP3)
The Web Content Analysis (Herring 2010) provides the project with real-world
data collected through Facebook's Ads Library API and the Twitter API. The Data
sets consist of adverts, announcements and messages from political parties
placed in these networks and related metadata like user interaction, links, the
period of publication, number of possible views, estimated gender of recipient
and locality of the display. It enables the project to analyze 1) the possible
target groups of the communication, and 2) design and message of the comm-
unication. The collected data will be depicted and analyzed with the help of
data visualizations. For each network, an individual visualization will be created
(cf. next section,Facebook and Twitter Map). The ndings can then be
compared against the ndings from the expert interviews who examine the
decision-making process behind the communication. Thus, both methods
complement each other in the effort of developing a comprehensive
understanding of how personal data inuences the political communication of
Swiss parties.
WebCA considers the information to be contained inside various types of
media fragments like links, exchanges or features. It is a methodological
plural paradigm and draws from methods like Computer-Mediated Discourse
Analysis, which has been applied to analyze forms of dialogue (chat rooms,
text messaging) (Herring 2010., 238) or Social Network Analysis, which has been
used to analyze political linking practices (Foot et al. 2006). It, therefore,
allows for the analysis of various types of information related to social network
data (e.g. text, image, metadata).
To dene the corpus of research a Web Sphere (Foot & Schneider 2002) is
created for every major party (SVP, FDP, CVP, GLP, SP, Grüne), consisting of
the party's presence on Facebook and Twitter. It is dened as a „hyperlinked set
of dynamically dened digital resources spanning multiple Web sites relevant
to a central theme or object“ (Ibid. 225) (cf. Fig. 10)
The project limits the data collection to Facebook and Twitter. Other
platforms like Instagram are excluded. Facebook’s and Twitter’s global reach
make them still the most dominant platforms for political communication
(Bradshaw & Howard 2019). The project can serve as a basis to explore other
platforms in future research.
Data Visualization (WP1 / WP2 / WP3)
The project develops three thematic visualizations (maps) throughout the
research process to gain new insights regarding actor roles and political
communication in networks:
1. TheActors Map(cf. Fig. 9, page 14)visualizes the involved
actors, their roles and relations, including the user, data brokers, parties,
2.3 Detailed Research
Plan
17/29
Predicting Voters
agencies, and the platforms. It visualizes how personal data is collected,
processed and utilized for political communication. The map focuses on
the decision making processes behind the adverts trough depicting key
ndings from the review and the expert interviews.
2. & 3. TheFacebookandTwitterMapsvisualize the political
communication of the parties inside the respective networks. The maps
are based on the data collected through the WebCA (cf. 2.3.4). It enables
the project to analyze the design and message of the communication
and to put it into relation with corresponding metadata about the target
groups (e.g. gender, locality, time). It furthermore enables the project
to analyze how the strategies of the parties differ in regards to frequency,
range of coverage, medium (image, text, video), message, nancial
expenses and how the users in the network interact with it.
Every visualization is divided into three parts commonly used in political
network visualizations (Pfeffer 2017):
1. Substance: Preparing the content/data, i.e. specic information on
the actors, data collection/utilization processes and political
communication.
2. Design: The process of transferring the substance (content/data) into
graphic elements. Here the visual means (e.g. color, size), as well as
interactivity, is dened (e.g. show/hide layers of information, zoom into
different dimensions of the data).
3. Algorithms: Development of the computer-based support necessary for
the processing and presentation of the data. They help, for example,
to reduce complexity by combining data or to position elements spatially
based on dened laws. To improve the visualization gradually the
project proposes a rapid prototyping approach conducting iterative
revision cycles of assessing needs, setting objectives, prototyping and
utilizing. (Baek et al. 2004). TheEthical Data Visualization
Workow(Hepworth 2019) will serve as a guideline to ensure that the
visualizations meet ethical standards. Other principles discussed,
such asvisualizing uncertainty(Griethe & Schumann 2005), are also
taken into account.
The different teams will develop the maps collaboratively (cf 2.3.7). The regular
exchange within the team ensures a development based on the team
member's needs. At the end of the project, the visualizations will be published
in the open-source documentation.
Fig. 10 Web Sphere consisting of
party afliated Facebook/Twitter
accounts.
Mother party Facebook
& Twi
er account
Relevant candidate Facebook
& Twi
er accounts
Other with the party linked Facebook
& Twi
er accounts which are
considered important (e.g. NGO’s or
newspapers)
Cantonal parties Facebook
& Twi
er accounts
2.3 Detailed Research
Plan
18/29
Predicting Voters
2.3.5 Work Packages (WP) & Milestones (M)
WP 1 Exploration Phase
A) Theory B) Data C) Design
1.1 Reviewing global/local
context on data-informed
political communication
1.2 Rening party/proling
actors
1.1 Reviewing data
brokers and platforms
1.2 Rening data broker/
platform actors
1.1 Reviewing local
agencies working with
political data
1.2 Rening agency actors
• Combining actors, describing characteristics and relationships
• Dene objectives of exploratory interviews, Actors Map, Facebook data col-
lection, Facebook-Map
1.3 Conducting explo-
rative expert interviews
1.4 Analyzing actors
based on interviews
1.5 Mapping actors (in
collaboration with C)
1.3 Setup Facebook API
access, review current
best practice, dening
Web Spheres
1.4 Collecting/pruning
Facebook data set
1.5 Processing Facebook
data set (in collaboration
with C)
1.3 Conceptualization
of visualizations (maps)
and team-user scenarios
based on the dened
objectives
1.4 Developing Actors
Map (in collaboration with
A)
1.5 Developing Facebook
Map (in collaboration with
B)
M1: Q4 2021 Explorative expert interviews conducted
M2: Q4 2021 Facebook API accessed, data set collected, pruned, and analyzed
M3: Q4 2021 Visualizations conceptualized, Actors Map and Facebook-Map
prototyped
• Based on conducted interviews, Actors Map, Facebook-Map: dene objecti-
ves of systematizing interviews, Twitter data collection, Twitter Map
WP2 Systematization of ndings
A) Theory B) Data C) Design
2.1 Identication and
enhancement of open
issues in regards to
involved actors
2.2 Conducting systema-
tizing expert interviews
2.3 Rening and
describing actors
2.4 Mapping actors (in
collaboration with C)
2.1 Identication and
enhancement of open
issues in regards to API
data source
2.2 Setup Twitter API
access
2.3 Collecting/pruning
Twitter data set
2.4 Processing Twitter
data set (in collaboration
with C)
2.1 Identication and
description of open
issues regarding the
maps
2.2 Elaboration of maps
2.3 Update and rene-
ment of Actors Map (in
collaboration with A)
2.4 Developing Twitter
Map (in collaboration with
B)
M4: Q4 2022 Systematizing expert interviews conducted
M5: Q4 2022 Twitter API accesses, data set collected, pruned, and analyzed
M6: Q4 2022 Actors Map updated and rened, Twitter Map prototyped
• Based on conducted systematizing interviews, updated Actors and Twitter
Map: dene objectives of theorising interviews, supplementary data set, vi-
sualization implementation
2.3 Detailed Research
Plan
19/29
Predicting Voters
WP3 Theorization of ndings
A) Theory B) Data C) Design
3.1 Conducting theorizing
expert interviews
3.2 Conclusive mapping
of actors (in collaboration
with C)
3.3 Evaluation and formu-
lation of the results of the
expert interviews
3.1 Renement of API
access
3.2 Collection/pruning of
supplementary
Facebook/Twitter data set
3.3 Processing supple-
mentary data set (in colla-
boration with C)
3.4 Evaluation and formu-
lation of data collection
process and APIs as data
sources
3.1 Finalization of Actors
Map (in collaboration with
A)
3.2 Finalization of
Facebook/Twitter Map (in
collaboration with B)
3.3 Implementation
of all maps, preparation
for publication
3.4 Evaluation and
formulation of the maps
and their functionality
M7: Q4 2023 Interview, data collection process, and maps evaluated and results
formulated, dissertations completed
WP4 Publication and dissemination
A) Theory B) Data C) Design
4.1 Completion of the nal research report (SNFS)
4.2 Dissemination of ndings in lectures at two international conferences (such as
IEEEVis and ACM SIGGRAPH), in two peer-reviewed journals (such as Communicati-
on Review Quarterly and Fibreculture), in the form of online open-source documen-
tation and the nal publication
4.3 publication of source code in an online-code repository
M8: Q4 2024 nal research report and conference concluded, dissemination of
ndings initiated
2.3 Detailed Research
Plan
20/29
Predicting Voters
2.3.6 Research Team
Organization
The research group unites 8 researchers from different elds and is divided into
three teams each consisting of an experienced team leader with a PhD in the
respective Field and a PhD candidate. One additional member is taking care of
direction and overall concept, another one of planning and coordination.
A) Theory Team
The theory team is responsible for the contextualization, organization and
conduction of the expert interviews. The members of the team arepolitical
scientistswith a specialization in political communication and propaganda
theories. The team members have a profound understanding of Switzerland's
political landscape.
B) Data Team
The data team is responsible for accessing, collecting, pruning, and storing the
network data (WebCA). The members of the team aredata scientistsfamiliar
with the most used languages for data analysis (R, Python, SQL, NodeJs,
others).
C) Design Team
The design team is responsible for the conduction of the visual prototyping
(Actors,Facebook, and Twitter Map). The members of the team aredesigners
with specializations in data visualizations, UX, and design research. They are
experts in data visualization tools like R and D3js and familiar with
programming.
2.3.5 Timetable
1
5
3
7
9
2
6
4
8
10
11
12
1
5
3
7
9
2
6
4
8
10
11
12
1
5
3
7
9
2
6
4
8
10
11
12
1
5
3
7
9
2
6
4
8
10
11
12
2021
2022
2023
2024
WP1
Exploration Phase
WP2 Systematization of
fi
ndings
WP3 Theorization of
fi
ndings
WP4 Publication and
dissemination
explorative
expert
interviews
Literature & Media Review
Literature & Media Review
Literature & Media Review
re
fining,
mapping
actors
re
fining,
mapping
actors
evaluation of
results
evaluation of
results
evaluation of
results
enhancement
of open
issues
enhancement
of open
issues
Finalization
of Actors-Map
Finalization of
Facebook/Twi
er
Maps
Implementation
of all maps
systematizing
expert
interviews
conclusive
mapping of
actors
theorising
expert
interviews
Completion of the
fi
nal research
report
Completion of the
fi
nal research
report
Completion of the
fi
nal research
report
Dissemination of
fi
ndings
Dissemination of
fi
ndings
Dissemination of
fi
ndings
publication of
source code
publication of
visualizations
Web Sphere,
Setup Facebook
API access
Setup Twi
er
API access
collection of
supplementary
data set
enhancement
of API data
source
Re
fi
nement of
API access
data
collection
data
collection
data
processing
data
processing
data
processing
Actors-Map
Update
Actors-Map
Facebook
Communica-
tion-Map
Theory TeamData Team
Design Team
Conceptualization
of visualizations
and team-user
scenarios
elaboration of
visualizations
Twi
er
Communica-
tion-Map
2.3 Detailed Research
Plan
21/29
Predicting Voters
Interdisciplinarity
During the literature and media review the teams collaboratively dene the
involved actors. A and C collaboratively develop the Actors Map based on
the conducted expert interviews. B and C collaboratively develop the
Facebook/Twitter Map based on the conducted WebCA. Every two to three
months of interdisciplinary work meetings with all teams will take place,
ensuring discussion about the process and the state of the research work.
2.3.7 Risks
The report by the SRF and personal exchange with journalists showed that
agencies and data brokers are potentially open for interviews. However,
it is important to acknowledge that some interview partners could not be willing
to share their knowledge or have no substantial knowledge about the
processes of data collection and utilization. The platforms most likely won’t
share deeper insights into their data processes, and neither are all political
actors willing to give an interview. But, it’s also important to note that
some parties (e.g. Grüne) are critical towards using personal data. I, therefore,
assume that they are interested in the ndings of the proposed research
and open for the exchange of information. General openness is expected from
the creator of the proling models who come from an academic context (e.g.
the Sinus Institute). Either way, enough time is calculated to expand the network
of possible interview partners. Additionally, the research does produce
valuable insights into the political communication of Swiss parties in social net-
works through its accompanying research methods. The project no only
relies on expert interviews as a data source but also conducts a web content
analysis that will provide knowledge about target groups, design, and
message of political adverts. A one-year preliminary study opens up important
contacts to actors even before the start of the project (cf. 2.5).
2.3 Detailed Research
Plan
22/29
Predicting Voters
2.3.7 Budget
All Prices are in Swiss Francs.
Material costs
Travel expenses interviews 22 à 100.— 2’200.—
Scientic articles in peer-reviewed
journals
2 à 3'000.— 6’000.—
Translations of scientic articles
(1 MS=50.–)
2 à 2'200.— 4’400.—
Conferences (travel costs,
each team once a year)
per person 500.— 12’000.—
Travel expenses for meetings per year 2'000.— 8’000.—
Transcription/analysis software for
interviews (e.g. Nvivo 12 Plus)
1'200.—
Total material costs 33’800.—
Final publication
Publishing services 5’000.—
Translation 8’000.—
Copy editing 3’000.—
Proofreading 1’000.—
Prepress 1’000.—
Digitization, Open Access publication 2’000.—
Image editing 1’000.—
Image rights 1’500.—
Print 10’000.—
Layout 15’000.—
Total nal publication 47’500.—
2.3 Detailed Research
Plan
23/29
Predicting Voters
Personnel costs
Direction 40% per year 40’000.— 160’000.—
Planning and coordination 40% 40’000.— 160’000.—
Team A leader 80% 80’000.— 320’000.—
Team A PhD candidate 100% 47'040.—
48'540.—
50'040.—
50'040.—
195’660.—
Team B leader 80% 80’000.— 320’000.—
Team B PhD candidate 10 0% 47'040.—
48'540.—
50'040.—
50'040.—
195’660.—
Team C leader 80% 80’000.— 320’000.—
Team C PhD candidate 100% 47'040.—
48'540.—
50'040.—
50'040.—
195’660.—
Total personnel costs 1'866’980.—
Total costs
Material costs 33’800.—
Final publication 47’500.—
Personnel costs 1'866’980.—
Total project costs 1'948’280.—
2.3 Detailed Research
Plan
24/29
Predicting Voters
2.4 Relevance & Impact
Scientic Relevance
The project provides crucial insights into the signicance of personal data for
political communication in Switzerland’s social networks. The scientic
community still lacks detailed knowledge abouthow Switzerland’s parties
collect and utilize personal data from potential voters to predict their
attitudes, motivations, and behaviors. Other studies, e.g. by the OII (cf
Bradshaw & Howard 2019), focus primarily on understanding the effect and
extent of digital propaganda and less onthe inuence of personal data
on political communicationin terms of design and message. Furthermore,
the involved actors, their relations to each other and their motivations
to work with personal data are treated marginally at best. The examination and
visualization of using personal data for political communication is for the
rst time the subject of an in-depth investigation of the example of Swiss ballot
meetings. This study not only looks at the macro-level of political com-
munication through an extensive analysis of political advertisements in social
networks but also at the micro-level through examining thedecision-
making process behind the data-informed campaigns. This knowledge is
considered essential for a differentiated debate about the use of personal
data for political means. The study contributes to discourses in the elds of
Political Science, Data Science (specically in regards to APIs as data sources),
Design Research (specically in regards to data visualizations) and the
Digital Humanities. The results are to be disseminated through lectures, articles,
the online open-source documentation and nal publication.
Broader Impact
As a result of the recent scandals around Cambridge Analytica and given the
much-cited 'over-complexity' of digital technology, there is an increasing
need for improved media literacy and critical ability. The project nourishes this
discussion by providing a comprehensive understanding of personal data
being used in political processes and thus makes a contribution to society
beyond the realm of science. The project raises the question of who has
the right to collect and utilize personal data with which intensions. In this way,
the project refers to one of the most important contemporary issues,
namely if we as a society want personal data to be a commodityand how this
interferes with basic rights like the protection of privacy. Against the
background of current practices and reform attempts in dealing with digital
media (e.g. decentralized networks, peer-2-peer communication), the research
results promote insights in a eld that has so far been treated highly
asymmetrically. The complexity of the matter, only allow small specialized
groups of programmers and global tech companies an understanding. Due to
the rapid technological development, there is hardly any room for
moments of reection in which society denes the social and political
requirements social networks have to meet. Various social actors can benet
from an understanding of the processes of political communication and
manipulation: State institutions concerning election observation, initiatives for
political transparency, social networks preventing political manipulation or
educational institutions in the respective elds.
2.4 Relevance & Impact
25/29
Predicting Voters
2.5 Next Steps
The following section describes the further procedure after the successful
completion of the Master's degree. The engagement with the political
dimension of the digital is reected in three elds of activity, namely academia,
practice and media art.
Academia
The project is submitted in smaller scale to the BFH Call for Proposals 2021 in
collaboration with Dr. Ulrich Fiedler (BFH TI) who assist as technological
expert. The preliminary study analyses Facebook adverts placed by the FDP in
regards to their target group, design and message. Simultaneously the research
plan is developed further to be submitted to the SNF. The SNF is seen as the
right funding partner because the project conducts basic research.
Sinergia, NRP or NCCR are considered appropriate programs to submit the
proposal because of it’s interdisciplinarity, actuality and relation to Switzerland.
Additionally, I became part of the project Participatory Knowledge Practices
in Analog and Digital Image Archives which has been submitted to the SNF. It
offers the opportunity to obtain a PhD in Digital Humanities (University of
Bern). My primary supervisor would be Prof Dr. T. Hodel while the three
co-applicants of the project (W. Leimgruber, P. Fornaro, U. Felsing) would share
the role of the second supervisor. In this context, U. Felsing and I develop a
joint article around the topic of data visualization. Additionally, I’m discussing a
joint article with my other mentor Katherine Hepworth.
Practice
I will include the diverse valuable experiences of the MA Design in my practice
(Studio Début Début) in the coming year. Concretely this means a revision
of the business plan that reects the increasingly scientic way of working and
the clientele. Additionally, this includes the development of a new service
offer dealing with the processing and visualization of data, specically in the
two contexts of archives and politics.
Media Art
Inuenced by political artists like Wachter/Jud or !Mediengruppe Bitnik, I
discovered art as a potential space for reection and mediation of my research
ndings. The Scripted Loopholes series is has been exhibited at the gallery
for contemporary art E-Werk in Freiburg during Regionale19. The series was also
proposed as a project at the Vilém Flusser Residency 2020, which would
take place between May to June 2020 at the University of the Arts Berlin. The
residency would provide the opportunity to develop the series further.
Academia
MA Project
SNF Project
PhD
BFH CfP 2021
First results
Application
Development
Archive
Project
PhD
Practice
Interaction
Design
Graphic
Design
„Data“
Design
Archive
Data
Political
Data
Media Art
Residency
2020
Vilém Flusser
Independent
Art Practice
Exhibitions
2.5 Next Steps
26/29
Predicting Voters
3. Documentation
A complete, annotated documentation of my master studies can be found
under: www.males.maxfrischknecht.ch
4. Declaration
I hereby declare that I wrote the present Master thesis myself and that I did not
use any other means than those listed, that I did not claim authorship of a text
and that I did not make unauthorized use of scientic texts or data.
Max Frischknecht, January 9th, 2020
5. Bibliography
Baek, Eun-Ok, Kursat Cagiltay, Elizabeth Boling, and Theodore Frick.
“User-Centered Design and Development.” In Handbook of Research on
Educational Communications and Technology, edited by David Jonassen
and Marcy Driscoll, 13. New York, NY, USA: Routledge, 2004.
Bogner, Alexander, Beate Littig, and Wolfgang Menz. Interviews mit Experten:
eine praxisorientierte Einführung. Qualitative Sozialforschung. Wiesbaden:
Springer VS, 2014.
Bradshaw, Samantha, and Philip N Howard. “Troops, Trolls and Troublemakers: A
Global Inventory of Organized Social Media Manipulation.” Oxford, UK:
Project on Computational Propaganda, December 2017.
Bradshaw, Samantha, and Philip N. Howard. “Challenging Truth and Trust: A
Global Inventory of Organized Social Media Manipulation.” Oxford, UK:
Project on Computational Propaganda, 2018.
Bradshaw, Samantha, and Philip N. Howard. “The Global Disinformation Order:
2019 Global Inventory of Organised Social Media Manipulation.” Oxford,
UK: Project on Computational Propaganda, 2019.
Cairo, Alberto. “Ethical Infographics: In Data Visualization, Journalism Meets
Engineering.” IRE Journal 37 (2014).
Chandler, Daniel, and Rod Munday. A Dictionary of Media and Communication.
1st ed. Oxford Paperback Reference. Oxford ; New York: Oxford University
Press, 2011.
Davies, Harry. “Ted Cruz Campaign Using Firm That Harvested Data on Millions
of Unwitting Facebook Users.” The Guardian, December 11, 2015, sec. US
news.
Dubois, Elizabeth, and Fenwick McKelvey. “Canada: Building Bot Typologies.” In
Computational Propaganda: Political Parties, Politicians and Political
Manipulation on Social Media. Oxford Studies in Digital Politics. New York,
NY, USA: Oxford University Press, 2019.
Drucker, Johanna. “Humanities Approaches to Graphical Display.” Digital
Humanities Quarterly 005, no. 1 (March 10, 2011).
Foot, Kirsten A., and Steven M. Schneider. “Online Action in Campaign 2000: An
Exploratory Analysis of the U.S. Political Web Sphere.” Journal of
Broadcasting & Electronic Media 46, no. 2 (June 2002): 222–44.
Foot, Kirsten A., Steven M. Schneider, Meghan Dougherty, Michael Xenos, and
Elena Larsen. “Analyzing Linking Practices: Candidate Sites in the 2002
3. Documentation
4. Declaration
5. Bibliography
27/29
Predicting Voters
US Electoral Web Sphere.” Journal of Computer-Mediated
Communication 8, no. 4 (June 23, 2006): 0–0.
Glowacki, Monika, Vidya Narayanan, Sam Maynard, Gustavo Hirsch, Bence
Kollanyi, Lisa-Maria
Neudert, Philip N Howard, and Thomas Lederer. “News and Political Information
Consumption in Mexico: Mapping the 2018 Mexican Presidential Election
on Twitter and Facebook.” Oxford, UK: Project on Computational
Propaganda, June 29, 2018.
Grassegger, Hannes. “Die Finkelstein Formel.” Das Magazin, no. 1/2 (January 12,
2019): 9–17.
Griethe, Henning, and Heidrun Schumann. “Visualizing Uncertainty for
Improved Decision Making,”, 4th International Conference on
Perspectives in Business Informatics Research (BIR’05), 2005.
Hedman, Freja, Fabian Sivnert, Bence Kollanyi, Vidya Narayanan, Lisa-Maria
Neudert, and Philip N. Howard. “News and Political Information
Consumption in Sweden: Mapping the 2018 Swedish General Election on
Twitter.” Oxford, UK: Project on Computational Propaganda, September 6,
2018.
Hepworth, Katherine. “Big Data Visualization: Promises & Pitfalls.”
Communication Design Quarterly Review 4, no. 4 (March 27, 2017): 7–19.
Hepworth, Katherine, and Christopher Church. “Racism in the Machine:
Visualization Ethics in Digital Humanities Projects.” Digital Humanities
Quarterly 12, no. 4 (2018).
Hepworth, Katherine. “Ethical Visualization.” https://github.com/kathep/ethics.
(October, 2019).
Herring, Susan C. “Web Content Analysis: Expanding the Paradigm.” In
International Handbook of Internet Research, edited by Jeremy
Hunsinger, Lisbeth Klastrup, and Matthew Allen, 233–49. Dordrecht:
Springer Netherlands, 2010.
Howard, Philip N., Gillian Bolsover, Bence Kollanyi, Samantha Bradshaw, and
Lisa-Maria Neudert. “Junk News and Bots during the U.S. Election: What
Were Michigan Voters Sharing Over Twitter?” Oxford, UK: Project on
Computational Propaganda, March 26, 2017.
Information Commissioner’s Ofce. “Democracy Disrupted? Personal
Information and Political Inuence,” July 11, 2018.
Information Commissioner’s Ofce. “Investigation into Theuse of Data
Analyticsin Political Campaigns: A Report to Parliament,” November 6,
2018.
King, Gary, Jennifer Pan, and Margaret E. Roberts. “How the Chinese
Government Fabricates Social Media Posts for Strategic Distraction, Not
Engaged Argument.” American Political Science Review 111, no. 3 (August
2017): 484–501.
Kosinski, Michal, David Stillwell, and Thore Graepel. “Private Traits and
Attributes Are Predictable from Digital Records of Human Behavior.”
Proceedings of the National Academy of Sciences 110, no. 15 (April 9,
2013): 5802–5.
5. Bibliography
28/29
Predicting Voters
Krogerus, Mikael, and Hannes Grassegger. “Ich habe nur gezeigt, dass es die
Bombe gibt.” Tages-Anzeiger, March 20, 2018, sec. International.
L’Ecuyer, Pierre, Patrick Maillé, Nicolás E. Stier-Moses, and Bruno Tufn.
“Non-Neutrality of Search Engines and Its Impact on Innovation.” Internet
Technology Letters 1, no. 1 (January 2018): e10.
Lacy, Stephen, Brendan R. Watson, Daniel Riffe, and Jennette Lovejoy. “Issues
and Best Practices in Content Analysis.” Journalism & Mass
Communication Quarterly 92, no. 4 (December 2015): 791–811.
Neudert, Lisa-Maria, Kollanyi Bence, and Philip N. Howard. “Junk News and Bots
during the German Parliamentary Election: What Are German Voters
Sharing over Twitter?” Oxford, UK: Project on Computational Propaganda,
September 19, 2017.
Newton, Casey. “Facebook Gave Spotify and Netix Access to Users’ Private
Messages.” The Verge, December 18, 2018.
MacEachren, Alan M., Anthony Robinson, Susan Hopper, Steven Gardner,
Robert Murray, Mark Gahegan, and Elisabeth Hetzler. “Visualizing
Geospatial Information Uncertainty: What We Know and What We Need to
Know.” Cartography and Geographic Information Science 32, no. 3
(January 2005): 139–60.
Machado, Caio, Beatriz Kira, Gustavo Hirsch, Nahema Marchal, Bence Kollanyi,
Philip N. Howard, Thomas Lederer, and Vlad Barash. “News and Political
Information Consumption in Brazil: Mapping the First Round of the 2018
Brazilian Presidential Election on Twitter.” Oxford, UK: Project on
Computational Propaganda, October 5, 2018.
McCrae, Robert R., and Paul T. Costa. “Personality Trait Structure as a Human
Universal.” American Psychologist 52, no. 5 (1997): 509–16.
Püntener, Daniela. “Hotspot: Meine Datenspur” Hotspot, n.d. https://www.srf.ch/
play/suche?query=hotspot%20datenspur
Pfeffer, Jürgen. “Visualization of Political Networks.” In The Oxford Handbook of
Political Networks. New York: Oxford University Press, 2017.
Popov, Vesselin. “How to Wield the Data-Driven Double-Edged Sword:
Navigating the Ethics of Psychological Proling with Big Data.” ZKM |
Zentrum für Kunst und Medien, Karlsruhe, Germany, April 12, 2017.
Reuter, Markus. “Social Bots: Was nicht erkannt werden kann, sollte nicht
reguliert werden.” netzpolitik.org (blog), May 9, 2019. https://netzpolitik.
org/2019/social-bots-was-nicht-erkannt-werden-kann-sollte-nicht-regu-
liert-werden/.
Ruch, André. “«Wahlkampf – Digital».” 10vor10. SRF Schweizer Radio und
Fernsehen, September 9, 2019. https://www.srf.ch/sendungen/10vor10/
wahlkampf-digital.
Sanovich, Sergey, Denis Stukal, and Joshua A. Tucker. “Turning the Virtual
Tables: Government Strategies for Addressing Online Opposition with an
Application to Russia.” Comparative Politics 50, no. 3 (April 1, 2018):
435–82.
5. Bibliography
29/29
Predicting Voters
Thomson, Judi, Elizabeth Hetzler, Alan MacEachren, Mark Gahegan, and Misha
Pavel. “A Typology for Visualizing Uncertainty.” edited by Robert F.
Erbacher, Jonathan C. Roberts, Matti T. Grohn, and Katy Borner, 146. San
Jose, CA, 2005.
Tufte, Edward R. Envisioning Information. 1st edition edition. Cheshire, Conn:
Graphics Press, 1990.
Tufte, Edward R. The Visual Display of Quantitative Information. 2nd edition
edition. Cheshire, Conn: Graphics Press, 2001.
Trice, Michael, and Liza Potts. “Building Dark Patterns into Platforms: How
GamerGate Perturbed Twitter’s User Experience.” Present Tense - A
Journal of Rhetoric in Society 6, no. 3 (January 21, 2018): 11.
Woolley, Samuel C., and Philip N. Howard. “Political Communication,
Computational Propaganda, and Autonomous Agents - Introduction.”
International Journal of Communication 10 (2016) (2016): 4882–4890.
Zeifman, Igal. “2015 Bot Trafc Report: Humans Take Back the Web, Bad Bots
Not Giving Any Ground | Imperva.” Imperva Blog (blog), December 9,
2015. https://www.imperva.com/blog/bot-trafc-report-2015/.
Zhdanova, Mariia, and Dariya Orlova. “Ukraine: External Threats and Internal
Challenges.” In Computational Propaganda: Political Parties, Politicians
and Political Manipulation on Social Media. Oxford Studies in Digital
Politics. New York, NY, USA: Oxford University Press, 2019.
5. Bibliography
