Digital News and Political Tweets in the Lower Austrian Municipal Elections: A Case Study on Digital Journalism and Political Communication

Abstract

In this paper, we study the problem of agenda setting by news media in relation to the political discourse by politicians at the time of local elections. We first evaluate the applicability of the agenda-setting theory against the theory of policy agenda building to determine the possible alternative directions for constructing a political agenda at the time of elections. Namely, we identify a non-linear interaction between news organizations, politicians, and the general public during the electoral campaign. This interaction, in turn, shapes the dynamic evolution of the public discourse concerning politics, and it is characterized by high sensitivity to initial conditions and non-linearity. Then, we attempt to identify the presence of an evolutionary trajectory of the political discourse in Lower Austria at the time of elections by observing whether, as the time of an election approaches, the interaction between news organizations and politicians flattens and becomes more linear without the news or the politicians causing the agenda of the other to be set accordingly. Finally, we provide a new methodology for identifying the topics contained in such an agenda so that empirical verification of the proposed hypothesis becomes possible.

1. Introduction

In this article, we study whether the type of relationship that exists between the content of digital news, on the one hand, and the content of public speeches by local politicians on social media, on the other, can inform us about the status of freedom of the press and freedom of speech, and the alignment thereof in a democratic system at the time of local elections. It is common to consider the freedom of the press and the freedom of speech as directly linked to the correct functioning of a democratic system. The theoretical reasoning that concerns this presumed linkage, though, dates back to an era far from the modern methods for delivering political content to the general electorate. The latter, today, though not at the time in which the concepts of freedom of press and expression were initially formulated, tends to occur in digital form. Consequently, some fundamental theories in journalism or political science have become subjectable to empirical validation, which has not always been possible in previous times. Specifically, it is now possible to test whether the presumed connection between the protection of a particular set of freedoms, notably, the freedom of expression and the freedom of the press, are functional requirements for a democratic system and whether they allow the representation of the population’s preferences at the level of decision making.

At the beginning of 2020, municipal elections tool place in Lower Austria. Elections are special moments in the life of a democracy because, within them, the private preferences of the population are expected to emerge and acquire a public stance (Feddersen and Pesendorfer 1997). This consideration is valid not only for national elections but also for local ones (Warshaw 2019). The process by which the private preferences of the electorate are shaped by the media, on the one hand, and the process by which politicians are selected, on the other, is, however, not necessarily linear; news may influence the choices of the electorate for some politicians, but politicians may also influence the choice of the electorates towards the selection of certain news content.

Researchers who have analyzed agenda-building and agenda-setting relationships between politics and the media during elections include, for example, Dumouchel (2022), who studied agenda-building efficiency during the 2015 federal elections in Canada. The results support the observation of non-linearity concerning agenda-building dynamics within the tension of daily issues and those pushed during media storm periods.

Another example can be found in the work of Pedro-Carañana et al. (2020), who analyzed discourses on Twitter in Colombia during the presidential elections. The study observed classical media and their influence on the Twitter discussion agenda. The results showed that the citizens dominated the campaign discussion and were less focused on the political program underneath. Furthermore, no statistical evidence was found between the presence of the candidate in social media coverage and the actual election results.

Within the Austrian domain, Seethaler and Melischek (2019) conducted their research concerning the use of Twitter for agenda building during the Austrian national elections in 2017. The results revealed that while Twitter can increase the parties’ agenda-building power, media agenda influence was not present. This situation only changed in the case of individual accounts of politicians who significantly aligned with the agenda presented by media outlets.

These studies focused on national elections. Hence, in this paper, we propose a method for studying the non-linear relationship between digital news and politicians’ digital communication, focusing on the events that led the institution of representation to be reinforced at the time of elections, specifically during local elections in 2020 in Lower Austria.

The article is structured as follows: First, we discuss the theoretical understanding of the literature concerning how digital news and digital communication by politicians function. This provides us with the a priori knowledge necessary to begin the research. Then, we address the two main theories concerning the relationship between media and politics in shaping the population’s political preferences. Finally, we apply the previously described concepts to define the expectations we would have concerning the alignment or misalignment of topics between digital news and digital communication by politicians. By collecting data on a sufficiently large corpus of digital news and political tweets by politicians, we can then empirically verify our research hypotheses. This, lastly, lets us test the expectation according to which digital news and digital communication should reinforce one another while simultaneously being independent of one another if, indeed, the institution of representation is present in a given political system.

2. Theoretical Framework

2.1. Defining Hypotheses from Theory

In order to develop the argument that we bring forward in this article, we have to construct an understanding of political systems, democracies, media, and political discourse and how these interact with one another. We do so by addressing the subject of the existence of political systems.

• There are polities.

The first assumption that we make is that there are political systems, such that some systems are political but not all. Political systems are often called polities: this word, plus its associated concept, is frequently employed to study politics empirically. For example, the concept of polity, or politics as a system, has been used to study the American (Erikson et al. 2002), German (Conradt and Langenbacher 2013), French (Safran 2015), and global political systems (Ougaard and Higgott 2002). More importantly, it has been used to study the relationship between politics, policy, and media in Austria (Zeitel-Bank 2017). For our purposes, it is, for now, sufficient to claim that polities exist, and that they can be studied under systems theory (King and Thornhill 2003). We call this hypothesis H_1.

2.

Polities are either democratic or not democratic, in a fuzzy manner.

The second assumption relates to a macro-attribute or characteristic of polities. Polities can be classified as democratic or not democratic, often referred to as autocratic or authoritarian (Tullock 2012). This statement may not be an abstract property of all political systems and may be bound to the analysis of a specific, limited period that corresponds to the recent era. We will not venture into the discussion as to whether democracy pre-existed the current era (Ward 2014) since that would be outside the scope of our research; however, it is worth mentioning that the consideration that a polity can be democratic or not democratic must be treated as an additional and independent assumption from the previous one since it does not necessarily derive from the definition of a polity as a political system. In our case, we assume that it makes sense to classify political systems according to the category of democratic or non-democratic polities and that there is some concrete method for doing so. The classification is, however, fuzzy (Gardashova and Aliyev 2014) because democracy can occur in a continuum of values rather than as a discrete typology. We call this hypothesis H_2.

3.

Democracy is a goal or tendency.

The third assumption corresponds to the theoretical understanding of the label we assign to political systems when we define them as democratic. In other words: what does it mean that a polity is democratic or not democratic? Political theory suggests that democracy can be seen as a goal or tendency towards which a polity orients itself (Gershman 2005). This, in turn, indicates that some of the things a political system does either promote or do not promote an invisible goal called democracy. These include programs for international police assistance (Marenin 1998); some others, maybe counter-intuitively, can also involve interventionist policy actions in foreign countries (Carothers 1991). There is some argument in the literature that the general tendency of humans to align their activities to benefit collectivity is as a result of natural selection. This idea states that democracy is tightly linked to trust and cooperation among individuals in a polity (Tov and Diener 2009), which leads us to call a system “democratic” when this cooperation exists between certain elements of it and non-democratic when cooperation does not exist. This points back to the old idea that cooperation between humans is an emergent property of biological evolution (Axelrod and Hamilton 1981), which in turn would also drive the evolution of political systems towards democracy at a higher level of abstraction (Cederman 2001). For our purposes, this assumption tells us that some particular goals, or the pursuit thereof, are democratic while others are not. The behavioral observable in this assumption is that, in a political system, some kind of alignment of the actions of the elements of the system should be observable if the system is democratic. We call this hypothesis H_3.

4.

Democracy requires a set of freedoms.

The fourth assumption is that, for a political system to be classified as democratic, a set of freedoms or liberties needs to exist. These freedoms are usually listed in the appropriate sections of a country’s constitution and comprise the constitutional foundations of the public sphere (Koltay 2019). The set of freedoms that have to do with information can be divided into two types: freedom of expression and freedom of the press.

Regarding freedom of expression, it has been noted that, in recent years, freedom of expression in liberal democracies has been declining, while the level of the autocracy of countries has simultaneously increased (Lührmann et al. 2019). This type of analysis is conducted by assuming synchronicity of the two variables; that is to say, it makes sense to measure them, as they are simultaneously observable in the same period. However, if we investigate whether freedom of expression is a necessary condition for democracy, we have to account for a specific time lag before the variation in the indicator associated with freedom of expression can affect the indicator associated with democracy, regardless of the metric we use. If we do not account for this, we are only providing evidence for the claim that the condition of necessity is not true; we have a reason to hypothesize that the same third invisible variable might be causing both the decrease in the freedom of expression and the decrease in the level of democracy. In the absence of this, we might also suggest that the two variables are not linearly independent and that, as such, the variation of one implies the variation in the other. The latter may well be the case; if we look at the datasets that are consulted to obtain the variables used in the study (ibid.), the V-Dem dataset1, for example, we notice that the indicator of democracy is constructed as a function (among other variables) of a variable called “Freedom of Expression”, which represents the associated concept. Thus, the hypothesis of linear independence of an independent and a dependent variable, which is a requirement for considerations of the causal relationship of the former on the latter, is violated when we consider the relationship between freedom of expression and democracy. This is akin to saying that “democracy” and “freedom of expression” are two different words to describe the same concept comprising two aspects “democracy” and “freedom of expression”.

The freedom of the press has a function that is analogous to freedom of expression. However, the former relates to the freedom of a specific subset of the general population: the population of journalists. The relationship between freedom of the press, politics, and democracy is not obvious. One might think that freedom of the press always defines a democratic system; contrary to this expectation, it has been noted that when a country becomes more democratic, this does not necessarily correspond to an increase in the freedom of the press (Sussman and Guida 2018). If this is true, that means that the condition of the necessity of freedom of the press for democracy is false. We will see in the next section how this problem might be solvable if we think that the functional independence between the press and politics, and not merely the capacity of the media to produce arbitrary content, is what corresponds to a system in which the freedom of the press is respected. We call this hypothesis H_4.

5.

Freedom is the absence of constrictions of the form “Thou Shalt Not”.

Because Freedom, with the upper-case F, is invisible, we need some behavioral observables that are tied to the presence of freedom, with lower-case f, or with the lack thereof. One conception of freedom postulates that liberty is a negative concept intended as the absence of interference by other persons on an individual’s behavior (Berlin 2017).

In the digital era, freedom of expression has acquired a new connotation, indicating the concrete possibility of generating and distributing information freely through digital channels. Because the cost of publishing and owning information has been nullified with the advent of the internet, the focus of freedom of expression in the digital era is moved from the capacity to distribute information without the physical interference of others to having the technological tools that allow the transfer of information to take place within a society digitally. The transfer of information to be free in a democratic society requires that democracy is promoted as a consequence of it and that autocracy is not promoted instead (Balkin 2004).

This is not to say that autocratic messages should not move over the internet. Instead, the argument is the following: if, as a causal consequence of the flow of certain pieces of information, representative democracy weakens, then that piece of information breaks freedom of expression. We will see in the later section why, based on this argument, we are allowed to study whether democracy worsens or not by examining the consequences of the generation and diffusion of certain information in terms of the loss of functional independence among the actors of the digital information space. In other words: if the generation and diffusion of a particular piece of information cause the system to become less representative of the citizenry, then that piece of information can be considered as going against the tendency of a political system towards democracy. We call this hypothesis H_5.

2.2. Media Agenda vs. Policies

The question then becomes, what exactly is meant by saying the actors of the information space should maintain functional independence from one another? The answer to this question emerges from studying the fundamental theories in journalism and their relationship to political communication.

In representative democracies, digital journalism plays a fundamental role in the diffusion of political messages to a population (Iosifidis and Wheeler 2018). Indeed, the belief in the existence of genuinely representative democracies, whose ontological nature is a subject of debate in political theory (Urbinati and Warren 2008), is carried forward by the narratives of political discourse that are pushed forward by digital journalism (Peters and Witschge 2015). That is to say, regardless of whether representative democracies are a thing of the universe, the belief in their existence, carried forward by digital journalism, produces observable behavioral consequences. The most notable of these is, of course, citizen participation in the political life of a country (Bakker and Paterson 2011). In addition, digital journalism via news recommender systems (Helberger 2019) can contribute to one of the theoretical requirements for the existence of representative democracy: full information about political matters by the citizen-elector (Hamilton 2015). If the citizens are not informed citizens, this is enough to claim that one of the fundamental assumptions of a democratic system is violated and, therefore, that a given political system whose citizens are not informed cannot function as a democratic system (Froomkin 2004). The literature on political communication is mixed on the subject; one of the basic assumptions for a democratic system to exist is a condition that is never and has never (Brown 1997) been satisfied: the condition of a fully and optimally informed citizenry. Other sciences would have discarded the concept and the associated theory as false, since it is a rather gross approximation of reality; however, in social sciences, they remain for the lack of a better alternative.

Instead, we treat the citizenry as capable of acquiring information about politics and having agency in expressing their political preferences. Two main theoretical approaches consider the role that news, citizenry, and politics play in decision making and on each other. These are the most common approaches to studying the problem of “which causes which” in the interrelationship between politics and news, in general, and policy making and the diffusion of information, on the other.

The first approach considers the news and the information contained therein as causally preceding politics and policy. This approach takes the name of agenda setting (Dearing and Rogers 1996). While in its original formulation, the theory appears old and related primarily to written journalism and television (McCombs and Shaw 1972), contemporary versions of it have been developed to cover the role that social media (Feezell 2018) and digital journalism (Nygaard 2020) can have in determining the subjects that matter for the political discussion of a community. If this approach is correct, it implies that whatever mechanism leads the news to orient itself toward specific topics will first be reflected in an increase in the information available to citizens on that topic; this, in turn, will end up affecting the subjects for discussion within the system of political decision making.

The opposite approach is that of policy agenda building (Nisbet 2008), which assumes the chronological and causal antecedence of politics over the news. Agenda building suggests that whatever mechanism leads to political decisions to be made leads the subjects discussed in politics to be subsequently reflected in the news produced by the media.

This process is considered by its supporters to be a regular mechanism for the functioning of modern democracy, in particular during election times (Dorantes y Aguilar 2014), while its detractors consider it akin to propaganda (Wilbur 2021). If we approach the subject as scientists and preserve our neutrality on the subject, we can realize that the two positions are equal: if we acknowledge that policy agenda building is a real thing, whether we consider it a typical method for the functioning of democracy or whether we consider it as propaganda used to manipulate the public opinion, we are nonetheless accepting the idea that policy precedes, chronologically and causally, the selection of topics that are covered by the media.

Both approaches suggest one thing: news and politics are at the two ends of an information pipeline that sees the citizen in the middle, and they function in such a manner that whichever affects the other has to create informed citizens first (see Figure 1). The informed citizen then either demands certain public policies to be adopted if the agenda-setting approach is correct or selects media content based on the political discussion in the political institutions if the policy agenda-building approach is correct. Regardless of which one of the two is a better description of how the world functions, they both require the citizens to acquire information for that information to be passed along this hypothesized information pipeline (see Figure 1).

The argument then continues by claiming that whenever the pipeline works better, then democracy also works better. It is frequently argued that democracy works if, and only if, citizens are informed (Aalberg and Curran 2012; Milner 2002). For this reason, we can think of informed citizenry as a desirable value towards which we should aim (Edwards 1998) and then evaluate the information technology we develop according to that fundamental value. This has been the case, e.g., for the role that the internet in general (Delli Carpini and Keeter 2002) and digital democracy in particular (Nisbet and Scheufele 2004) plays in promoting informed citizenry.

If, however, we assume that such a thing as an informed citizenry can and does indeed exist, it then makes sense to study the relationship between the organizations that center around politics, the political parties, and the organizations that center around information: the media in general and the newspapers in particular.

2.3. Methodologies to Study Agenda Building and Agenda Setting in Digital Journalism

The increasing amount of data and information available, including the velocity, present a constant challenge to journalists during their research (Perreault and Ferrucci 2020). Hence, tools and approaches based on (semi-)automated content analysis have gained more and more attention among the journalist research community (Flaounas et al. 2013). Boumans and Trilling (2016) provide an overview of the spectrum of tools, ranging from deductive to more inductive approaches. A classical approach was pursued by Pérez-Díaz et al. (2020) in their work focusing on the agenda-building process of digital news media. Building on an existing set of deductive codes, the authors categorized, classified, and compared trending Twitter topics with digital news topics of major Spanish news outlets. Another deductive study—using a dictionary-based approach—was conducted by Ceron et al. (2016), who analyzed two polarizing political topics in Italy via statistical and sentiment analysis to investigate the first-level and second-level agenda-setting effects of social media. Moving towards inductive approaches, one example can be seen in the work of Guo (2012), who explored social network analysis in the overall context of agenda-setting research. Following this inductive direction, Jacobi et al. (2016) discussed in detail the suitability and applicability of topic modeling approaches in digital journalism. An example of its application in the domain of agenda building can be found in Prytkova et al. (2021), who studied the representation of Russia in digital news from a Ukrainian and Kazakhstanian perspective via a topic model for news agenda extraction. In our paper, we combine two of the aforementioned approaches, i.e., the topic model and network analysis, focusing in particular on text-based networks.

3. Elections in Lower Austria and the Electoral Participation of the Resident Population

We can now proceed to discuss the socio-political context within which the case study that we selected took place. The most recent municipal elections in Lower Austria, which are the subject of this article, were conducted on the 26 January 2020. The elections took place in 570 Gemeinden of the Federal State of Lower Austria and saw the active participation of five political parties, their associated lists, and other independent candidates. Out of the 1,669,944 residents of Lower Austria, the number of persons eligible to vote was 1,480,968, accounting for 88.68% of the total population. In total 972,530 of those with the right to vote cast ballots, corresponding to a voter turnout of 65.66%. As a baseline for comparison, it is worth mentioning that the voter turnout in Austria for the elections of the European Parliament in 2014 and 2019 was, respectively, 45.39% and 59.80% (European Parliament 2019). The voter turnout in Austria for the Austrian national parliament’s elections was 80.00% and 75.59%, respectively, in 2017 and 2019 (International IDEA 2018). This indicates that the turnout of Lower Austrian residents in municipal elections is located somewhere between that of the European Parliament and that of the Austrian Parliament if we assume that the statistics for Lower Austrian residents are representative of the statistics for the general Austrian population.

Out of the total votes submitted, 98.34% were deemed valid and contributed to the result (Land Niederösterreich 2020). The five political parties that competed in the elections are the five parties that are also represented in the Austrian National Council. These are The People’s Party of Austria (Österreichische Volkspartei—ÖVP), the Social-Democratic Party of Austria (Sozialdemokratische Partei Österreichs—SPÖ), the Freedom Party of Austria (Freiheitliche Partei Österreichs—FPÖ), the Greens (Die Grünen—Grüne), and the New Austria and Liberal Forum (Das Neue Österreich und Liberales Forum—NEOS). In addition to the independent candidates, the same parties had previously competed in the Lower Austrian municipal election in 2015; therefore, a diachronic comparison of the two elections is possible.

Table 1. Results of the 2020 and 2015 municipal elections in Lower Austria.

Party	2020 Elections		2015 Elections		Difference
	Votes	Percentage	Votes	Percentage	Votes	Percentage
ÖVP and allies	503,969	52.69	491,211	50.26	+12,758	+2.43
SPÖ and allies	265,524	27.76	303,054	31.01	−37,530	−3.25
FPÖ and allies	55,562	5.81	75,850	7.76	−20,288	−1.95
GRÜNE and allies	56,448	5.90	43,934	4.50	+12,514	+1.40
NEOS and allies	12,066	1.26	8661	0.89	+3405	+0.37
Independent lists	62,833	6.57	54,664	5.59	+8169	+0.98

Source: (Land Niederösterreich 2020).

shows the results of the two elections.

The recent elections saw the coalition’s victory led by the ÖVP, followed by the SPÖ and its allies. Together, these two coalitions accounted for 80.45% of the preferences expressed by the electorate of Lower Austria. At the time of writing, the ÖVP is in a government coalition with GRÜNE, while SPÖ is in opposition, as are all other parties.

The distribution of preferences expressed by the electors in Lower Austria represents the number of seats given to political parties in the National Council of Austria. Figure 2 shows the relationship between the two variables.

These distributions are highly correlated, having a value of p = 0.93. This suggests that the preferences expressed by the electorate in Lower Austria can generally be considered representative of the preferences of the larger Austrian electorate. This is important because it means that, to a certain degree, we can shift the level of analysis from the federal level (Austria) to the local level (Lower Austria) to compare the preferences of the electorate between the two levels. This would not have been the case if, e.g., the electorate of Lower Austria had a predominant preference for candidates in non-marked or independent lists over those that are associated with more established political parties. In that case, we would not have been able to shift the level of analysis from federal to local and vice-versa due to the lack of representativeness of the latter over the former.

In the following sections, we will discuss how this consideration is useful in comparing political opinions and political communication as expressed at the two levels of analysis.

3.1. The Landscape of Digital Communication at the Time of the 2020 Municipal Elections

At the time of the elections, two digital channels played a primary role in the diffusion of information related to the proposed candidates and the relevant political events associated with them. These were the digital press, discussed in the following section, and Twitter, which we will discuss later.

Regarding the digital press, the local and federal press covered the elections extensively in the weeks immediately preceding the elections and for a few days afterward. In Austria, the landscape of the printed press is characterized by the presence of two prominent newspapers that publish at a federal level but also have regional publications. These are followed by a more extensive set of regional newspapers, which are more contained in terms of both readership and coverage. The latter collection of newspapers frequently comprises a single federal state or Länder (Trappel 2007) territorially. Among the players at the federal level, Kronen Zeitung is referred to as Krone, traditionally associated with political conservatism and populism (Faber and Unger 2008). Since 2004, Krone has been publishing regional versions of its newspaper, which are distributed free of charge, initially in Vienna and later in various federal states, including Lower Austria. Krone, however, continues to remain, in essence, a national newspaper, since it covers primarily Viennese or pan-Austrian topics and events.

Another national newspaper is Der Standard, which is typically compared with the German newspaper Die Welt and the Spanish El Mundo due to its pro-socialist positions (García-Avilés et al. 2014). Der Standard has a lower coverage than Krone, printing and selling only around 10% of the copies of its larger competitor as of 2011 (Gobbato 2014). This newspaper counts as national or federal for the purpose of this research, which makes it possible to compare it to Krone in terms of territorial coverage.

In addition to the two nationwide newspapers, we also consider a set of newspapers that are more relevant for studying the diffusion of information in Lower Austria. The first of these is Niederösterreichische Nachrichten (NÖN/NOEN—Lower Austrian News), which is published in St. Pölten, the regional capital of Lower Austria. According to its self-reports, NOEN has a readership of 480,000 readers among Lower Austrian residents for the period 2015/2016. This puts it slightly ahead of Krone in terms of reach, since the latter had 474,300 readers in the same period (Fleck 2016). We do not make claims as to the political affiliation of NOEN; instead, in compliance with the literature (Kelly et al. 2004), we assume only that its information has a strong local bias or focus.

The next local newspaper that we consider is Tips.at, whose motto reads “total.regional” [sic] and which claims that “regionality is the recipe for success”. Tips does not have a federal edition; instead, it has regional editions that pertain to some of the regions of Austria: Linz, Innviertel, Hausruckviertel, Traunviertel, Mühlviertel, and more importantly for our purposes, Niederösterreich (Lower Austria). These regions do not correspond exactly to the local administration’s names as per Austria’s administrative law. For example, Innviertel and Mühlviertel comprise two different areas of the same federal state, i.e., Upper Austria. However, in the case of the regional edition of Tips for Lower Austria, there is perfect correspondence between the administrative unit and the territorial coverage of the newspaper.

The final digital newspaper that we consider in our study is Meinbezirk, whose title translates in English as “My District”. Meinbezirk is another regional newspaper that is owned by the company Regionalmedien Austria AG, which also operates the newspaper’s web portal. The Lower Austrian edition of Regionalmedien Austria’s newspaper was distributed to 721,492 readers in 2020, according to Austrian Edition Control (Österreichische Auflagenkontrolle 2020), which would put it above all others in terms of distribution. However, in that particular report, the figures include both the digital and the paper version, and the digital version refers to an electronic copy of the physical newspaper, not to the number of those accessing the web pages of the digital newspaper. The publication, in either case, remains relevant for the readership in Lower Austria and has coverage that is good enough to justify its inclusion in our research. An overview of the presented news outlets and their coverage can be seen in

Table 2. Summary of the key features of the five newspapers analyzed and selected as data sources for our study.

Acronym	Full Name	Centrality	Coverage	Political Position	News Collected
KRONE	Kronen Zeitung	Federal	818,859 copies sold in 2011	Conservatist	2674
DS	Der Standard	Federal	72,693 copies sold in 2011	Socialist	4247
NOEN	Niederösterreichische Nachrichten	Local	480,000 in Lower Austria	Unknown	5477
TIPS	Tips.at	Local	Unknown	Unknown	1957
MEIN	MeinBezirk	Local	721,492 readers in Lower Austria as stated by its publisher	Unknown	1287

Source: Elaborated by the authors, based on the information contained in this section.

.

Of course, these are only some of the newspapers that exist, and there are other digital media that are present and operate in Austria in general and in Lower Austria in particular. For now, though, we assume that the newspapers we have selected represent the set of digital newspapers published in the region. Since this research does not focus on analyzing the complexity of the media landscape but rather attempts to support a certain statement concerning the interaction of digital newspapers and political communication over social media empirically, we consider the assumption of representativeness of our sample over the entire set of digital newspapers as sufficient for our purpose.

3.2. The Usage of the Tweets of Politicians to Study Political Communication

The general appropriateness of studying social media messages to learn about political communication in a country is commonly acknowledged. The relationship between the two is covered extensively in undergraduate students in communication sciences (Bruns et al. 2015; Harvey 2013), and we accept that social media and news can therefore affect one another. Among the various social media outlets, the most common media in the literature on digital communication and politics is Twitter. This is because of its saliency at the time of the most recent American elections as an instrument for political communication by the presidential candidates (Chen et al. 2021; Gainous and Wagner 2013) but also because it later became traditional as a tool for political communications in countries other than the United States, including in European contexts (Bolgov et al. 2018; Maurer 2019). Indeed, Twitter has also been used to study political elections in Austria, which is particularly relevant for our research (Kušen and Strembeck 2018; Seethaler and Melischek 2019).

The usage of Twitter by politicians presents regularities that allow scholars to study the function of the underlying phenomenon of political communication based on the functions that these regularities perform. At the time of elections, it has been observed that there is a general tendency by political parties to use Twitter to communicate to the population about their electoral platform, among other usages (Jungherr 2016). However, individual politicians prefer to use Twitter to express their own political agenda and to politicize and mobilize the general population into participating in electoral events such as rallies and public speeches (López-Meri et al. 2017).

The same presidential candidates, upon being elected, then change the pattern of usage of Twitter and do not necessarily employ it as a preferential channel for communication with the electorate. For example, it appears that at times of strong social and health crises, as the COVID-19 pandemic has proven to be, the usage of traditional channels such as the television is more likely to effectively transfer a message from a lead politician to the general population (Teufel et al. 2020). This may be a peculiarity of the German-speaking world, though, since other studies claim that the opposite is true and that the usage of Twitter by politicians is indeed a conveyor of useful public health information (Rufai and Bunce 2020). Regardless of this particular point, the literature agrees that politicians use Twitter more frequently during elections. During this period, they use it to convey messages that pertain to the upcoming vote and present content meant to influence the electorate in their favor.

The alignment between politicians’ tweets and news media content has been noted previously (Posegga and Jungherr 2019). However, the literature is not unanimous in claiming a preferential direction in the relationship between political tweets and newspaper content. While some claim that the news picks up topics from tweets by politicians shown by a certain number of stories containing the same topic or hashtag (Shapiro and Hemphill 2017), other studies suggest that tweets follow the newspaper articles, at least for some particular topics such as climate change (Su and Borah 2019). This suggests the relevance of studying the functions that Twitter messages play in terms of the creation and diffusion of a topic for political communication at the time of elections and also in terms of the relationship that those topics have with those that are, simultaneously, transmitted to the general population by the newspapers of the same country.

4. Research Questions and Hypotheses

After elaborating on the theoretical premises according to which this study was conducted, we can now formalize the research questions and hypotheses in a language akin to propositional logic. This will let us identify a hypothesis on the relationship between media and politics that we can subject to empirical testing, or rather, whose corresponding null hypothesis we can attempt to falsify. This study has five hypotheses based on the theoretical foundations in Section 2.1.

$H_1$: There are polities.

$H_2$: Polities are democratic or not democratic, and their positioning along the axis of democracy–autocracy is continuous and fuzzy rather than discrete.

$H_3$: Democracy is a goal or a tendency of a system. It is not a state or configuration of the system; it is also irrelevant whether a fully democratic system can exist or not in the universe. The tendency or orientation towards democracy is real, whether democracy in itself is real or not.

$I_1=(H_1\wedge H_2\wedge H_3):$ Polities either tend or do not tend toward democracy, and they do so in a continuous and fuzzy rather than discrete manner.

$H_4:$ Democracy requires a set of freedoms, which comprise at least the freedom of the press, the freedom of expression, and the freedom of speech. For our purposes, the freedom of the press is distinct from the freedom of expression and speech, since it concerns different actors. Freedom of speech and expression generally relate to the whole population and also to the elected representative of the citizen-electors. Freedom of the press refers to the media in general and digital newspapers in particular.

$H_5:$ Freedom is the absence of constrictions of the form “Thou Shalt Not”, related to the production of written text in the case of the freedom of the press and to the production of political speech in the case of the freedom of speech or expression.

$I_2=\left(H_4\wedge H_5\right)$: The tendency towards democracy requires, and therefore implies, the absence of certain constraints concerning the production of written texts by the press and political speeches by politicians.

$I_1\wedge I_2\leftrightarrow P\to Q:$ If a polity tends toward democracy $\left(P\right)$, then the constraints related to the production of texts and speeches ($Q$) are marginal. If we can prove that this implication is correct, and if we also had a magical tool that would allow us to prove that the premise $P$ is valid in some concrete political system, then the modus ponens would allow us to deduce that the consequence $Q$ is also valid for that same political system. For this reason, it is useful for us to prove or disprove the validity of this implication.

We can test whether the implication is substantiated by applying the second law of De Morgan. We thus can negate the entire term $I_1\wedge I_2\leftrightarrow P\to Q$ and thus construct an empirically testable null hypothesis for our study. The negation of the left term $\left(I_1\wedge I_2\right)$ produces the following:

$$\neg \left(I_1\wedge I_2\right)\leftrightarrow \left(\neg I_1\right)\vee \left(\neg I_2\right)$$

The negation of the right term $\left(P\to Q\right)$ produces the following:

$$\neg \left(P\to Q\right)\leftrightarrow \left(Q\to \neg P\right)$$

By combining the two expressions, we can derive the null hypothesis $H_0$, which we employ in our study:

$$H_0:\left(\neg I_1\right)\vee \left(\neg I_2\right)\leftrightarrow \left(Q\to \neg P\right)$$

Converted into natural language, the null hypothesis reads as follows:

Either polities tend or do not tend towards democracy in a fuzzy manner $\left(\neg I_1\right)$, or the tendency towards democracy does not require the absence of constraints on the press and expression ($\neg I_2)$, which can be true if and only if, at the same time, democracy requires the absence of certain constraints on the diffusion of information, and if, by removing those constraints, the polity does not tend towards democracy.

4.1. Data

The data used in this study were collected during the period of municipal elections in Lower Austria in 2020, with the day of the election being 26 January 2020. The collection started for both the news data and the Twitter data on 1 January 2020 and lasted until 15 February 2020. This collection period was selected to cover relevant conversations and news reports before and after election day. Overall, 15,642 news items (see Table 2 for sources) and about 3000 tweets2 were collected. In order to use these data for the sake of this study, DeepL3—a professional deep learning-based automated translation service—was used to translate all German tweets and news into English. It has previously been shown by the National Research Council Canada (NRC) that the translation quality of automated services such as Google Translate and DeepL is sufficient for their use in research and analytics contexts (Bhardwaj et al. 2020).

4.2. Methodology for Topic Extraction

In order to subject the research hypotheses to empirical testing, we needed a method for extracting the themes or topics contained in the data we collected. Because the corpus of text was too large to use qualitative methods for coding the texts into topics, automated methods were necessary.

The process of extracting topics from larger text corpora is commonly referred to as topic modeling. Classical approaches in topic modeling, such as Latent Dirichlet Allocation (LDA)-based approaches (Blei et al. 2003), apply statistical probability models to group words from provided documents, where each document is seen as a combination of topics, with each word belonging to a topic. LDA-based approaches are common in many domains, including, e.g., the field of journalism (Von Nordheim et al. 2018). As the field of topic modeling develops further, new approaches are introduced, which take the automated detection of latent themes beyond pure bag-of-words-based approaches by leveraging the potential of network analysis and, thus, graph theory. The term (social) “network analysis” is usually connected to the analysis of relationships between people, see, e.g., Knoke and Yang (2020); however, it extends beyond this application to a variety of domains (see i.e., Pitoski et al. 2021; Lampoltshammer and Wiegand 2015). This flexibility can also be leveraged to apply network analysis to the application domain of journalism. Specifically, it aims to show the network-based relationships between words (i.e., co-occurrences). This introduces several advantages in comparison to the previously mentioned LDA-based approach. One example is that retaining the context of the extracted words and their relationship is still possible, which cannot be done via a bag-of-words approach. In addition, while there are restrictions concerning the amount of required data for a sustainable analysis with LDA, these restrictions do not apply to network-based approaches. Finally, the network approach also allows the application of research results from the domain of community detection (Traag et al. 2019), which can then be used to identify topics within the generated network. The R-based software library for topic modeling used for this study is called “textnets”4 and is based on the work of Bail (2016). This library allows for the construction of two-mode networks out of co-occurrences for text analyses. While the decision of the right balance between smaller co-occurrences (pair is found, e.g., only once) or larger co-occurrences (i.e., pair is found multiple times) from which to extract topics is mainly based on the experience of the researchers conducting the analysis, a common way of setting a cut-off threshold is the so-called “elbow method”. This method is used to optimize the analysis towards the point of diminishing returns of having larger samples. A common application scenario for this approach, e.g., can be found in finding the right number of clusters within a dataset (Syakur et al. 2018). This logic was applied to both datasets, i.e., the digital news dataset and the Twitter dataset. Based on this approach, the actual network is calculated and exported to Gephi (Bastian et al. 2009). Then, in the first step, layout algorithms are executed on the network to increase visibility and, thus, interpretability by changing the visual representation of the network. In this case, we applied ForceAtlas2 (Jacomy et al. 2014) and Yifan Hu (2005). Afterward, the Leiden algorithm (Traag et al. 2019) was applied to detect clusters of co-occurring word pairs. Then, the average degree of the respective network, i.e., Twitter and digital news, was calculated. This enabled the further separation of the co-occurrences. Here, the concept of entropy was followed, and thus the assumption that the lower the frequency of interconnections within the network, the higher the importance associated with the remaining interconnections. The networks were then further cleaned and resulted in the following extracted “mini networks”, representing samples of the included topics with the respective Twitter data and the digital news data. These topics were then used to connect back to the original data, i.e., the original text of the tweets (489) and the news items (1046) associated with these topics to better interpret the underlying meaning of the extracted topics.

4.3. Annotation of the Topics

We then proceeded to validate the empirical claim that the two sets of topics, pertaining respectively to the digital newspapers and the political tweets, are related to but are also independent of one another. To perform the validation, we first computed the Cartesian product between the two sets of topics, which in this case were represented as a list of words contained in each of the clusters indicated above. Because the clusters were 10 for the digital news and 10 for the political tweets, this comprised a Cartesian product containing 100 elements, corresponding to all possible pair-wise combinations of the elements of the two sets. Then, we asked 50 human non-experts for each pair to assign a similarity score to the 2-tuples corresponding to all elements of the cartesian product of the two sets. The score was given on a Likert scale that ranged from 0 to 4, as per

Table 3. The Likert scale that human annotators used to score the pair-wise relatedness and similarity between the topics.

Score	Description
0—Unrelated	The two lists do not mean the same thing and are not on the same topic.
1—Slightly related	The two lists describe dissimilar concepts, ideas, and actions but may relate to one another in a longer document on the same topic.
2—Somewhat related but not similar	The two lists have dissimilar meanings but share concepts, ideas, and actions that are related to one another
3—Somewhat similar	The two lists share many of the same important ideas, concepts, or actions but include slightly different details.
4—Very similar	The two lists have very similar meanings, and the most important ideas, concepts, or actions contained in one are represented in the other

below.

This scale was selected following the literature on the human annotation of semantic similarity of texts (Jurgens et al. 2014). The annotators were paid for their work and were not aware of the objectives or claims that constituted the subject of the research. The question that they were presented with was “How similar are these two topics?” without any indication concerning the topics’ origin or the methods for their extraction; further, there was no indication concerning the expected or a priori distribution of similarity measures across the pairs of topics. Finally, to account for the known cognitive bias according to which the order in which information is presented affects judgment (Bratton 2010), and by knowing that information presented earlier in a sentence weighs more than information presented later, we ordered the words in a topic according to their ascending alphabetical order.

Table 4. The topics extracted from the corpus of digital news and the corpus of digital tweets in the application of our methodology.

Topics in the Digital News

Topics in the Political Tweets

(n1) ankara, coast guard, libya, refugee, tripoli, tunisia (n2) afghanistan, alliance, kabul, pakistan, taliban, troops, usa (n3) escalation, guard, killing, iraq, militia, soleimani, usa (n4) euro, progression, relief (n5) birth rate, family, korea, society, south korea, woman (n6) household, food, food waste, waste (n7) birth, home birth, hospital, midwife (n8) bougainville, guinea, panguna, petition, referendum (n9) degree, master, master craftsman, title (n10) answer, disease, pandemic, pathogen, protection, question, vaccination, virus

(t1) abolition, headscarf ban, pension (t2) agitation, alma, solidarity (t3) casino case, head, law section, pilnacek, supervisor (t4) cancer, vaccination, vacuum (t5) ibiza, servants, tweets (t6) context, eu commission, face recognition software, nikischerak, photo, signature, situation (t7) climate crisis, generation, government programme, preamble, programme (t8) fashion, idea, obbligation, parliamentary club, receipt, ticket (t9) cargo bike, federation, food, morning, read, wheel (t10) company, goal, reduction, task force

represents the list of extracted topics, with their tokens sorted according to their alphabetical order, which was submitted to the human annotators.

4.4. Results and Discussion

Finally, we extracted a weighted network (see Figure 3) out of the ranked scores on the pair-wise similarity of topics, as assigned by the annotators. The preferences of the annotators were first aggregated by calculating the arithmetic mean for each pair of topics; then, we filtered out all pairs for which the mean was lower than 1. The reason for this choice is that if the average score of a pair of topics is lower than 1, we deem that pair to be unqualified for the label of “slight relation”, which is associated with a measure of similarity of one. All other pairs of topics were used to form the edges of an undirected weighted graph, where the weight corresponded to the arithmetic mean calculated earlier.

The graph is sparse, albeit not strongly, possessing just 45% of the total possible edges for a fully connected graph of that type. While the graph has an order $\left|V\right|=20$, the maximum theoretical size $\left|E\right|$ is not ${{\displaystyle \sum }}_{i=1}^{\left|V\right|-1}i=190$ because there cannot exist edges between two vertices belonging to independent subsets. The maximum theoretical size, instead, is $\max \left(\left|E\right|\right)=10\times 10=100$, which is only possible if all topics are related to all others. If that had been observed in our graph, this would have been strongly suspicious and would have indicated that the methodology we used was flawed; instead, a more appropriate sparsity that corresponds to slightly less than half of that of a complete graph is more realistic. On the other hand, the graph might have also been empty if no edges between topics were retained after filtering by weight. If that had been the case, this would have indicated that each topic that is present in the tweets is unrelated to all topics that are present in the news, and vice versa. If that had been the case, this would have immediately confirmed the validity of the null hypothesis $H_0$.

This is not the case, though, which leaves some room for interpretation, as will be discussed shortly. First, though, we can inspect some salient features of this graph to check whether its structure is theoretically consistent with some of our expectations. The first consideration we can make is that each topic in each subgraph is connected to at least some of the topics in the other subgraph. The visible exception concerns the vertex t8, corresponding to a weird mixture of words, as seen in Table 4. The topics were extracted from the corpus of digital news and the corpus of digital tweets in the application of our methodology above. The weight of the edges that connected this topic before filtering averaged 0.806, with a minimum value of 0.58 corresponding to (n2, t8) and a maximum value of 0.92 corresponding to (n4, t8). The latter value is just short of the minimum threshold required to pass the filter, which corresponds to a weight of 1. The topic t8, however, seems to us to be largely unrelated to the topic n4. This is valid, with the exception of the idea that the tokens “euro” and “tickets” may be related to the idea of a price to be paid for purchases, and the tokens “parliamentary club” might be associated with the idea of “relief”, since the latter could be a policy of the former. It seems, however, that t8 is indeed an odd topic, which therefore justifies its classification as unrelated to anything else.

Secondly, we can notice how n7 results are exclusively connected to one topic, t4. The connection between these topics is reasonable because the token “hospital” contained in n7 is understandably related to the tokens “cancer” and “vaccination” contained in t4, since they all relate to the concept of health.

Thirdly, we can observe that the highest weight observable in the graph corresponds to the edge between n10 and t4. This is understandable for the same reason that is valid in the case of the pair (n7, t4). Topic n10 and topic t4 relate to health and therefore to one another, and n10 relates to health more closely than topic n7 because more tokens are associated with that idea.

Finally, it should be noted we do not have any edges whose weight is greater than three. Because a score of three or four would indicate not only relatedness but also similarity, we can infer that no topics in the news are similar to any topics in the political tweets, though they are related to some of them, and vice versa. The consequence is that we can accept the functional independence of the news and the political tweets regarding selecting the specific topic they cover. At the same time, though, we can acknowledge that these topics, while being functionally independent among the two sets, maintain enough relatedness among them.

The interpretation that we can give of this observation is that the null hypothesis is rejected. That is to say, it appears that freedom of expression for the politicians, on the one hand, and freedom of the press for the digital newspapers, on the other, is being maintained in the period that approximates the municipal elections in Lower Austria. The theoretical generalization that we can give to this statement is that the five hypotheses upon which $H_0$ was formulated are not falsified. That is to say, there are political systems that can tend or not tend towards democracy, and tending towards democracy requires, simultaneously, the functional independence of the press and politics but the alignment of the content that they express. This is intended in the sense that, simultaneously but independently, the topics contained in digital newspapers and the digital communication by politicians become aligned. This occurs, however, without either of them taking over and determining the content of the other.

5. Conclusions

In this article, we discussed the problem of non-linear interaction between digital news and digital communication conducted by politicians at the time of local elections. We began by formalizing a set of concepts and formal propositions concerning the way in which the political system and the system of digital news are expected to interact with one another if a system is democratic. We subjected to empirical verification the hypothesis according to which, if the system is democratic, then digital political communication and digital news should, independently but concurrently, propose topics that are aligned with one another. The argument supported is that if the system favors the institution of representation of the political preferences of the population, it should then be observable that digital news and digital communication by politicians are aligned with one another without, simultaneously, any of the two determining the content of the other. We tested this hypothesis on the content of the digital news that was published in Lower Austria at the time of the local elections and by comparing the topics extracted from that news with those derived from the tweets by politicians. This let us develop a network of topics simultaneously covered by the two datasets. Finally, by observing that the network of topics contained in the two corpora presented certain characteristics of similarity, we then argued that the institute of representation during the political elections in Lower Austria is reinforced, and not damaged, by the interaction between digital news and digital communication by politicians.

Some limitations of this research are the partial representation of digital newspapers in the corpus that we collected, such that a subsequent study may complement ours by providing more extensive coverage of the Austrian newspapers published in 2020. Methodological limitations may also arise from the attempt to extract topics utilizing our topic modeling technique first and the methodology for analyzing the network of topics second. Other techniques for topic extraction may result in different results than the ones we showed here, though we expect the semantic meaning of the extracted topics to be similar. Lastly, the decision to limit the network of topics to a threshold of 1, such that all edges with a weight under that threshold do not show, was, to some extent, arbitrary. While it makes sense to do so given the choice of the Likert scale that we used, a scale that is supported by the literature on semantic similarity, there may be sensitivity effects that change the layout of the network in the case that a different threshold is provided for the cut.

Possible continuation of this work may include categorizing the digital news and tweets collected according to their time index to search whether the alignment between topics is stronger closer to the day of the election. This, in turn, would allow the study of the dynamic evolution of the network of topics and their interaction as an extension of the synchronic analysis that we provided here. Another possible extension consists of the application of the methodology presented within this article to study digital communication during elections in federal states other than those in Lower Austria, as the current approach and results have a local focus within one federal state only. This, in turn, would allow the verification of whether any specific cultural or political reasons may explain the alignment of topics we observed; this would not be a general property of a political and media system and would rather be specific only to the Austrian case. Finally, a comparative study of theoretical results from previous works and their practical implications for processes and procedures in media and political communication, combined with deep-dive interviews with practitioners in both domains, could reveal interesting insights within this science-to-practice translation.