The Influence of Social Media Usage as a News Source on Trust in Politics

(1)

The Influence of Social Media Usage as a News

Source on Trust in Politics

Derek Heek

10741267

Bachelor Thesis Information Science Supervisor: Mr. ir. A.M. Stolwijk

University of Amsterdam December 14, 2017

Abstract

In the current digital age, the role of social media as a news source keeps growing. This paper probes to find out how the public’s trust in pol-itics is affected by this development. A structural model is designed based on several hypotheses, substantiated by an extensive literature study. This model contains eleven latent variables: trust in political content in or on (1) Facebook, (2) Twitter, (3) news on television, (4) current affairs pro-grammes on television, (5) talkshows on television, (6) news on the radio, (7) newspapers and (8) journals. Then someone’s (9) social trust and (10) social and cultural background are taken into account. Finally, the dependent latent variable is (11) trust in politics. Using a questionnaire (N=131), those variables were measured based on indicators found in lit-erature. Using structural equation modelling, the relations between the latent variables were measured. Trust in current affairs programmes and social trust showed a positive, significant relation to trust in politics. As well did trust in Facebook on trust in TV news, current affairs programmes and newspapers. Trust in Twitter showed a positive and significant rela-tion to trust in radio news and journals. Trust in radio news itself showed a positive and significant relation to social trust. No significant relation between trust in social media as a news source for political content and trust in politics is detected.

keywords: Trust, Dutch Politics, Social Media, News, Structural

Equa-tion Modelling

(2)

2.4 Measurement Model . . . 16 3 Method 17 3.1 The Survey . . . 17 3.1.1 Structure . . . 17 3.1.2 Likert-scale questions . . . 18 3.1.3 Procedure . . . 19 3.2 Data processing . . . 20 4 Results 20 4.1 Measurements . . . 22 4.1.1 Reliability . . . 22 4.1.2 Validity . . . 23 4.2 SEM Analysis . . . 23 4.2.1 Model Adjustments . . . 25 4.3 Final Model . . . 26 5 Discussion 28 5.1 Recommendations . . . 30 References 31 A Survey Questions 35

(3)

1 Introduction

In an era where social media are playing a more and more prominent role, the implications on society of these developments should be investigated. In 2016, the term post-truth has been elected word of the year. It turned out that this adverb showed an increased use in combination with the noun politics, espe-cially during political events like the Brexit referendum and the US elections (Oxford-Dictionaries, 2016). Post-truth is defined by Oxford-Dictionaries (nd) as "Relating to or denoting circumstances in which objective facts are less in-fluential in shaping public opinion than appeals to emotion and personal belief". According to Casper Grathwol, director of Oxford Dictionaries, post-truth is defined as "...fuelled by the rise of social media as a news source and a growing distrust of facts offered up by the establishment".

The development of people using social media as a news source instead of the traditional media is notable, since it can be argued that this is less reliable than verified news sources. This development also invites politicians for a dif-ferent campaign approach, with potentially more possibilities, because they are now capable of directly target their audiences or the public in general, without interference of the press.

1.1 Uprise of Social Media

At the time of writing, 81% of the US citizens has an account on social media, whereas in 2008, this was 24% (Statista, 2017). The worldwide number of people with an account on a social media platform is estimated on 1.96 billion and the growth is not expected to stop, since this number is expected to grow to 2.5 billion in 2018. Facebook has by far the biggest market share with around 1.5 billion profiles. In 2016, 79% of US adults used Facebook, whereas the second most social medium, Instagram, was used by 32% of the same population (Greenwood et al., 2016). Those numbers are all huge and therefore the word uprise is being used in this research to describe the rapid developments. 1.1.1 Filter Bubble

The concept of the filter bubble has been described by Pariser (2011) as a personalised version of the World-wide-web for everyone. Every internet user is basically in his or her own bubble, in which he or she sees content that is in line with what that person likes, what that person’s interest are etc. It can be stated that all content that a user gets to see, confirms his er hers perception of the world.

So this filter bubble is basically an algorithm that predicts what you want to see, based on what you liked, watched or searched for before. To the biggest websites in the world (Facebook, Google, Yahoo etc.) this is a smart business model, since they can provide ’free’ services to their users by making money from using their user’s data by for example targeted advertising. Also the more time a user spends on for example Facebook, the less time that user can spend

(4)

on let’s say Twitter. So the more interesting a website or app is, the worse it is for its concurrent, which again enables the company to generate more money. 1.1.2 Pillarisation

Even though it can be argued that it is unethical what companies with this filter bubble strategy do, the concept of living in a filter bubble is not new. Pillarisation has played an important role in the second half of the 20th century in the Netherlands for example. The implications of living in a divided segment of society might be more extreme than living in a filter bubble, but the concept is basically the same. There are however some new dynamics within the filter bubble according to Pariser (2011); everyone is in his or her own bubble, which means you are the only one in a unique bubble. Next to that, it is probable that a regular user is not aware that he or she is in a filter bubble. Certain content might be filtered from a Facebook user’s newsfeed unasked. This could for example be left-winged political content if that user reads or watches a lot of right-winged content online. Finally, when a person is on the internet, he or she automatically enters his or hers personal bubble, it is not a choice. The moment a user goes to a website or watches certain content, the filter is being adjusted.

1.2 The Paradox of Western Open Society

Main characteristics of Western open society are freedom of speech and a liberal democracy. Those values are being under pressure because of themselves, as a result of the above described uprise of social media. It has never been easier to share information or an opinion than it is now to such a large audience. The problem with social media is that the provided information does not have to be true, it is not being validated, while it gets published. This means that anybody can share whatever they want, whether this is true or not. People who share the same worldview, might see this content, because of the filter bubble, and agree with it, because they want it to be true, what brings us back to post-truth politics.

Fake news is a relevant topic in the post-truth era, since it could be used to influence the public opinion with facts that are not true, for example Donald Trump denying climate change, so-called alternative facts, in other words: lies or misinformation. Fake news can be spread for seemingly innocent reasons, like gaining money from advertising with click-bait titles on social media, of which potential effects are unintended, but could certainly be substantial.

The real danger is in actors that have intentions to undermine the Western democracy by spreading fake news. This has been a hot topic during the US elections in 2016, where, at the time of writing, investigations are still going on to what extend Russia has been interfering in those elections. In elections that followed in Europe, Russia has again been accused of interference by spreading misinformation. However, it is rather hard to really prove this interference and especially whether the interference has been effective. Whether the interference

(5)

actually took place and had effect or not, it is important to realise that this low-threshold option is available for superpowers like Russia. This also means that smaller state actors, with no nuclear capabilities or big armies, can join the digital war. Smaller states can now make a significant difference on the power balance in the world, depending on their cyber capabilities.

In the context of post-truth politics combined with the uprising of social me-dia, Fish (2016) warns for an illusory democracy. This is still a liberal democ-racy, but politicians can gain power by doing a simple trick. This trick is to influence the public opinion in such a way that people will vote for them, which is basically the practical form of post-truth politics, as described in the first section of this research. As the title of this subsection already says, this is very paradoxical in a liberal democracy, since it can be argued that the government will no longer be a reflection of what the people really want, which is actually the purpose of this democracy.

1.3 Relevance

In the current post-truth, social media, digital age, it could be argued that the functionality of democracy is at stake. This research aims to expose the actual effects of the technological developments regarding social media. If social media turn out to indeed affect trust in politics, this paper should should raise awareness among the public and politicians. It might be possible that the traditional democracy, as it exists right, cannot exist in the modern world. This means it might be expedient to rearrange democracy in a way it fits today’s issues.

1.4 Aim of this Research

As described above, social media putting pressure on the foundations of Western society and the role of politicians in this new media is changing, because their influence is growing. This paper probes to find out how this replacement of conventional media is actually affecting the public’s trust in politics. All these elements are put together in the following research question:

"To what extend is the use of social media relative to conventional media as a news source affecting the trust of the Dutch public in its national politics?" First of all, the concept of trust in the given context needs to be defined. There is no standardised unit to measure trust, so an alternative way has to be found to make claims about it. Social media is a broad term, but the most important aspect of it is that users can create and share their own content within a social network and can interact with other people online. The focus in this research will be on Facebook and Twitter. To social media can also be referred to as new media, whereas old or traditional media refers to conventional media. Television, radio, (online) newspapers and journals are the four main conventional media

(6)

channels considered in this research, which will be further elaborated in the related work section. The most import characteristic of the old media is that there is a one-way communication to the public and the content in it is created by objective journalists. The Dutch public are all Dutch people who are qualified to vote during elections. Finally, politics includes all politicians who are part of the parliament and the functioning of the government as a whole.

Outline of this paper. The following section will go more in depth on the concept of trust. From that, a structural model will be proposed, based on a va-riety of hypotheses, which subsequently leads to its corresponding measurement model. The method section describes how those hypotheses in the measure-ment model were tested, which leads to a refinemeasure-ment of the proposed model. The paper concludes with a discussion on the research.

2 Related Work

In order to answer the research question, it is necessary to define the concept of trust. Section 2.1 discusses this concept and also why it is expected that social media has an effect on trust in politics.

2.1 Trust

Nevejan (2007) describes two reasons why trust exists in human behaviour, namely to survive or to improve someone’s well-being. According to Kohring & Matthes (2007), our modern society would not be able to function without trust. So perhaps without realising it, trust is a fundamental part of daily life. Nevejan further describes in her dissertation four dimensions that determine the amount of trust an individual would have in someone else. Those factors are relation (i.e. how well do you know this person, who is that person), place (i.e. how far away is this person (physically)), time (i.e. when was the engagement and how long did it take)and action (can it be done). Each dimension consists out of four factors which all have a certain intensity, which combined measure that dimension and give a certain configuration in which a individual’s trust is now. All those dimensions with their factors are put into one framework, the YUTPA-framework.

2.1.1 Changing dimensions

The reason to assume that social media is affecting trust in politics, is because the perception of content read on social media is expected to be different than content read or heard from conventional media. A politician might evoke more trust by posting content on social media, which he or she controls himself, rather than being reported by a conventional medium. It is expected that the configuration within the YUTPA-framework is different on political content read on either an old or new medium. For instance, social media has the potential

(7)

to make one’s perception of distance to the person, which in this case would be a politician, closer. When for example Donald Trump puts out a tweet on Twitter, it might feel to a follower of him as if he talks to him or her personally. Next to that, a user is able to directly respond to a message that is put out on social media.

Furthermore, the conventional media are starting to use social media more and more as a news source themselves (Broersma & Graham, 2012). The role of the press is partly a controlling one, to check whether politicians keep their promises to the people who elected them (Hudock, 1999). Instead of monitoring and controlling politicians, politicians themselves can now put out anything they want through any social medium. Next to that, a trend is visible in which right-winged populists, like again for example Donald Trump, are trivialising grounded institutions, where the conventional media base their information on. Recent examples of this trend are Trump denying climate change and calling judges fake. By trying to reduce trust in science and the legal system, next to calling the conventional media "fake news" all the time, the trust in conventional media could reduce as a whole. If that would result in more trust in the politician himself, he could fill this gap up by generating own news and facts on social media. This is in line with the problem described in the introduction section as well, regarding the post-truth age.

Based on above reasoning, it is expected that social media affects trust in pol-itics, since politicians can make themselves feel closer to social media users, interaction can become faster and the perceived distance to a politician re-duces. This might also improve the relation, as perceived by voters, between a politician and a voter. However, while this keeps speculating on a existing framework, it forms enough reason to investigate whether social media indeed affect trust in politics.

2.2 Structural Model Proposal

In order to actually measure the trust in politics, a structural model is proposed in figure 1. Five latent variables are included in this proposal. The relations between those variables are based on hypotheses, which are substantiated in the following subsections. Trust in politics is the dependent variable in this model. Eventually, in order to measure the latent variables in the structural model, a measurement model will be proposed. This model contains the indicators, based on which the latent variable will be measured. Those indicators are based on previous research and will be measured by conducting a questionnaire.

(8)

Figure 1: Proposal of a Structural Model on Trust in Politics

2.2.1 Culture and Social Background

Leigh (2005) defined three types of factors that influence a voters voting be-haviour, i.e. his or hers partisan choice. Those are individual, local and national factors. Individual factors include whether a person is rich or poor and whether the voter is young or older. He also found a difference between men and women, but this gap has closed over time. Local factors include whether a person lives in a rich or more ethnically diverse or unequal neighbourhood. Lastly, it mat-ters whether a voter is a native born or a foreign born. Even though Leigh’s research is not explicitly investigating trust in politics, it can be argued that one’s partisan choice can be associated with one’s trust in politics.

Indicators taken into account in this research of cultural and social back-ground are: CSB1: age, CSB2: relation to NL, CSB3: ethical diversity neighbourhood, CSB4a: contentment on living circumstances, CSB4b: con-tentment on income, CSB4c: concon-tentment on social life, CSB5: type of neigh-bourhood and CSB6: political orientation. Those indicators are based on the above described work of Leigh (2005). Note that CSB2 is relation to NL and not native or foreign born. This is because it is probed to not offend people with the question.

(9)

2.2.2 Social Trust

Social trust, i.e. to what extend do people trust other people, is measured by the General Social Survey (GSS) (nd) by asking participants the following question: "Generally speaking, would you say that most people can be trusted, or that you can’t be too careful in dealing with people?", which people could answer with: can trust, cannot trust, depends, don’t know, no answer, not applicable. The NSD & the ESS (2013) asked the same question, but on an 11-point scale. Nevertheless, Miller & Mitamura (2003) point out that the question was initially used as part of a five question index and that despite a lot of researches used this question, it has never really been validated. They also asked this same question on a 7-point scale, in the form of: Do you think most people can be trusted?, which gave other results than when they provided the question as the GSS asked it.

Brehm & Rahn (1997) designed a similar model with political trust as de-pendent variable. One of their indede-pendent variables was interpersonal trust, which is considered the same as social trust. Next to the ’can people be trusted’ question as described before, they used two more questions about whether peo-ple will take advantage of you when then get the chance or try to be fair, and whether people are helpful or only look out for themselves. Because these are again scale questions with basically two questions in one to answer, the ques-tions have been adjusted, so they can be answered on a agree-disagree scale. This lead to the following indicators of social trust (ST):

ST1: people will profit from me if they get the chance, ST2: most people are helpful towards other people, ST3: most people only take care for themselves and ST4: most people can be trusted.

2.2.3 Trust in Conventional Media

Gaziano & McGrath (1986) provide 12 dimensions on which media credibility can be measured. It can be argued that perceived credibility is a measure of one’s trust in a medium. Meyer (1988) reduced these factors to only 5, namely believability, biased, completeness, accuracy and trustworthiness. He also argues that those dimensions almost measure the same thing, but by asking for those multiple dimensions, that might just slightly differ, the measurement of this medium’s credibility will be far more accurate. This 5-question set reduced by Meyer, has been validated by West (1994).

However, the dimensions believability (fair vs. unfair) and trustworthiness (can be trusted vs. cannot be trusted) are so similar, especially when being translated to Dutch, that it is decided to merge those two. This is inline with the research of Rimmer & Weaver (1987) who also use four dimensions and where believability is not taken into account, only trustworthiness is.

Flanagin & Metzger (2000) conducted a research on whether people perceive sources on the internet as credible as old media sources. They found that their participants considered internet sources as credible as the conventional

(10)

media, except for the newspaper, which they thought was more credible. There are two interesting things about this article; (1) it was conducted in the early days of the internet. In their time of writing, the internet had an estimated amount of 130 million users, whereas social media themselves have over a billion now, as described in the introduction. It is also not news on social media they researched, since that was not really a big thing in 2000. (2) Furthermore, they used the five dimensions as given by Meyer (1988). Respondents had to rate every dimension on not at all to extremely on a 7-point scale for every medium. They researched this for four information types (news, references, entertainment and commercial), but this thesis will limit to news sources only.

In almost all researches that have been read during the literature study on this topic, conventional media are considered to be television, radio and newspapers. In this research, TV has been split into TV-news, current affairs programmes and talkshows, since in all different types, political content is being presented or discussed in a different way. The platforms are different in such a way that the four dimensions as described above could be filled in differently. For example, the content as discussed in a talkshow could be considered more subjective than on the news. Also a program on current affairs can go more in depth on political matters, which could result in more perceived completeness of the medium. For radio, news programmes have been chosen, since the news on radio is mostly discussed in an interactive way.

Next to newspapers, journals have been taken into account. It is expected that a minority of people actually read political content in those journals, but for those who do, it is expected to have an effect on their trust in politics. Online newspapers are considered a variable of conventional media as well. It is chosen to use the same latent variable for online and offline newspapers, since most of the online ones are extensions of the offline papers. However, online content can be updated real-time, while a daily newspaper only gets published once a day. Next to that, online newspaper-websites can use the richer-get-richer effect, which basically means that popular news items are chosen by amount of hits and also displayed on that popularity order. Anyway, it is not expected that this affects the contents of the items, which is why for now there is no difference made between online and offline newspapers.

Taken all of the above into account, the indicators trustworthiness (a), com-pleteness (b), bias (c) and accuracy (d) are used in this research, as the indica-tors of the different media. The latent variable trust in conventional media splits into trust in CM1: news on television, CM2: current affairs programmes on television, CM3: talkshows on television, CM4: news programmes on the ra-dio, CM5: newspapers and CM6: journals. The frequency of the medium use will be taken into account, however not as an indicator for trust in the medium, as suggested by Newton & Saris (2003), but to see if people use it on what scale and if that makes a difference on how they judge the medium. Time-spent is not used, because it requires serious reflection from a respondent to answer how much time on a weekly day he or she watched television for example, with .5 hour intervals. Next to that, Newton & Saris (2003) themselves point out that people spend a lot of time reading tabloids, but this does not mean they trust

(11)

them. Rimmer & Weaver (1987) also claim that the frequency someone uses a medium is not correlated with the credibility of this medium.

2.2.4 Trust in Social Media

The latent variable trust in social media in the structural model is basically the one of which the existence is probed to be prove in this paper. Based on the context as described in the introduction, it is hypothesised this entity does exist now.

Because social media are considered also a form of media, it will be measured with the same indicators as described in the above section. In this way, both media-types (social and conventional) are measured the same way. The variable trust in social media is split up into trust in SM1: Facebook and SM2: Twitter. The reason Facebook has been chosen, is because according to Barthel et al. (2015), 39% of the Facebook users consider it "an important way of keeping up with the news". To 4% of their respondents it even was the most important way. Also, Van der Veer et al. (2017) found that it is the second most used social medium in the Netherlands with 10.4 million users, of which 7.5 million users on daily base. The most used social medium in the Netherlands is Whatsapp, but this is mostly used for inter-personal communication only, so it is left out of the scope of this research. Number 3, 4 and 5 in their research are respectively Youtube, LinkedIn and Instagram. Youtube is not taken into account, because Baumgartner & Morris (2010) already found that only a small percentage of their respondents (17%) used Youtube more than once a week as a news source. Even though there is political content on it, in this research it is considered more of an entertainment website. LinkedIn is popular, but mostly used for work-and career-related content. It is not expected that people would use LinkedIn to form a opinion on politics. Instagram is becoming more and more popular, but it is image based and expected not to be political-related, because it does not offer the most practical platform to discuss content. However, politicians could be active on it to express their political ideas. Despite decreasing popularity, Twitter is still considered relevant.

It is notable that Twitter is not in the top five of Van der Veer et al. (2017). It however was in it until 2016. In 2017, it still has 2.6 million Dutch users, of which 871.000 on daily basis. Twitter is also a medium that is discussed a lot in conventional media.

2.2.5 Trust in Politics

A documented question development by the European Social Survey (ESS) pro-vides questions on measuring trust in political institutions (Curtice et al., 2003). Those questions could be indicators of the latent variable trust in politics. Only the question on the United Nations has been left out in this research, since it is not necessarily relevant for Dutch national politics. The European Union, however, has been left in, since this is a big topic in Dutch politics. Participants had to fill in on a 1-10 scale what their personal trust is in the given institutions,

(12)

e.g. in political parties, politicians etc. Schneider (2017) also used the ESS as a source for her survey. Her questions differ slightly from the ESS final decided survey, but she validated her model, so questions from her work will be included in the survey for this research.

The indicators of trust in politics in this research are trust in: TP1: the current Dutch government, TP2: Dutch House of Representatives and Dutch Senate, TP3: political parties, TP4: politicians, TP5: local council, TP6: the Dutch legal system, TP7: the police and TP8: the European Union.

The following hypotheses with their sub-hypotheses have been stated, based on the assumed relations between all the described latent variables so far to trust in politics:

Null Hypothesis 1 A Dutch voter’s trust in Dutch politics has no relation with his or her trust in political content on Facebook (H1a) or Twitter (H1b) There are several reasons to believe the relation between trust in social media and trust in politics exists and therefore to believe that hypotheses H1a and H1b can be rejected. Brewer et al. (2013) for example found that exposure to news coverage of satire, which is usually shared on social media, has an effect on someone’s knowledge, opinion and political trust. Gil de Zúñiga et al. (2012) support those findings, because they found that the use of social media to find information and engaging in political discussion networks, predicts one’s political participatory behaviour.

If trust in politics would be expressed in terms of voter turnout, Valenzuela et al. (2009) found an positive relation between intensity of Facebook use and political and civil engagement. Effing et al. (2011) showed that Dutch politicians in the national elections of 2010 with a higher presence on social media, got relatively more votes. This is in line with people using Facebook on a higher intensity, since they would read more about or from politicians that generate more content, instead of politicians generating no content at all, or not reading any content that a politician shares or generates.

Sherchan et al. (2013) argue that the government could use social media in order to provide its services to citizens. but in order to do so, trust of the public in those media is needed. Therefore they conducted a literature review on trust in (web-based) social networks. Their work is another indicator that the relation between trust in social media and trust in politics exists.

Null Hypothesis 2 A Dutch voter’s trust in Dutch politics has no relation with his or her trust in political content on news on television (H2a), current affairs programmes on television (H2b), talkshows on television (H2c), news on the radio (H2d), newspapers (H2e) or journals (H2f )

It is assumed that the relations between trust in conventional media and trust in politics exists, because for the past century, this was one of the only ways to inform yourself as a citizen and form an opinion on politics. So if the old media were a voter’s (main) source of information on politics, those media can directly influence the public’s trust in politics. Therefore it might be desirable for the

(13)

conventional media to be objective. However, there are still politicians claiming that the public television channels (NPO in Dutch) are more left-winged. This could reduce trust in the right-winged Dutch politics. Next to that, as described in the filter bubble section, during the pillarisation in the Netherlands, every pillar had their own newspaper or TV-channel they watched. However, whether the conventional media have been objective or not, it is very likely that they influenced trust in politics over time.

Null Hypothesis 3 A Dutch voter’s trust in Dutch politics has no relation with his or her social trust

Zmerli & Newton (2008) found that social trust is positively correlated with trust in political institutions and satisfaction with democracy. This is in line with the results of Schyns & Koop (2010), who found that an higher individual’s social trust results in less distrust in politics. Based on this way of thinking, H3 could be rejected.

Null Hypothesis 4 A Dutch voter’s trust in Dutch politics has no relation with his or her cultural and social background

Substantiation for hypothesis H4 comes from the work of Leigh (2005) as de-scribed in the beginning of section 2.2.1.

2.3 Remaining Relations in the Structural Model

In this subsection, the remaining hypotheses will be stated and substantiated, which form the other relations in the structural model.

2.3.1 Towards Trust in Conventional Media

In the very early 2000’s, when social media hardly did not exist yet, Kiousis (2001) researched the news credibility perception of newspapers, online news and televised news. He found that people in general are already sceptical of these media channels. He also found that interpersonal communication on televised news has a negative influence on the perception of news credibility. It is notable that this does not account for interpersonal communication about news read in the newspaper. Anyhow, social media offer a platform to communicate about the news, so they can reinforce the distrust in conventional media.

Lazarsfeld et al. (1948) found that voters tend to expose themselves to the propaganda of their own party more than to the propaganda of the other party. Please note that this research dates from 1948, but it is still relevant, because of the filter bubble as described in section 1.1.1. This filter bubble might make this factor even stronger, because it is designed to have people use an app for a longer time, by showing content that is in line with the user’s view on the world. This means that a user mainly gets to read stuff, he already believes is true, which will make the user even more convicted of its perception of the world.

(14)

Marchi (2012) found that teenagers prefer opinionated news over objective news, which could respectively be considered news read on social media over news read in conventional media. Note that this research was conducted among teenagers who are not allowed to vote yet. Still, it is showing a trend and it is relevant for a upcoming generation.

The dynamics between old and new media are interesting to look at as well. As Broersma & Graham (2012) found, social media are actually being used as a news source by old media. So if for instance a politician tweets something, newspapers might use this to write an article about. Also, because so many people have access to social media, they can share images or video’s just minutes or even seconds after for example a terror attack took place. For old media, this forms a source they need to use, in order to provide news coverage first. A side effect of this development is that the shared images on conventional media might become more violent and shocking.

The expected relation between a voter’s cultural and social background to trust in conventional media within the proposed structural model is not so relevant for the outcome of this research. However, the data to check if the relation exists, will become available. Therefore there will not be spend too much attention too it. When the results are available, it will be checked whether the null hypotheses 6, 8, 10, 12, 14, and 16 can be rejected. If they can, this relation is interesting to investigate in an other research.

Null Hypothesis 5 A Dutch voter’s trust in political related news on television has no relation with his or her trust in political content on Facebook (H5a) or Twitter (H5b)

Null Hypothesis 6 A Dutch voter’s trust in political related news on television has no relation with his or her social and cultural background

Null Hypothesis 7 A Dutch voter’s trust in political related content in current affairs programmes on television has no relation with his or her trust in political content on Facebook (H7a) or Twitter (H7b)

Null Hypothesis 8 A Dutch voter’s trust in political related content in cur-rent affairs programmes on television has no relation with his or her social and cultural background

Null Hypothesis 9 A Dutch voter’s trust in political related content discussed in talkshows on television has no relation with his or her trust in political content on Facebook (H9a) or Twitter (H9b)

Null Hypothesis 10 A Dutch voter’s trust in political related content dis-cussed in talkshows on television has no relation with his or her social and cultural background

Null Hypothesis 11 A Dutch voter’s trust in political related news on the ra-dio has no relation with his or her trust in political content on Facebook (H11a) or Twitter (H11b)

(15)

Null Hypothesis 12 A Dutch voter’s trust in political related news on the radio has no relation with his or her social and cultural background

Null Hypothesis 13 A Dutch voter’s trust in political related news in the newspaper has no relation with his or her trust in political content on Face-book (H13a) or Twitter (H13b)

Null Hypothesis 14 A Dutch voter’s trust in political related news in the newspaper has no relation with his or her social and cultural background Null Hypothesis 15 A Dutch voter’s trust in political related content in jour-nals has no relation with his or her trust in political content on Facebook (H15a) or Twitter (H15b)

Null Hypothesis 16 A Dutch voter’s trust in political related content in jour-nals has no relation with his or her social and cultural background

2.3.2 Towards Trust in Social Media

A recent study by Newman & Fletcher (2017) shows that people with high income disagree more to the proposition that social media does a good job in distinguishing facts from fiction than people with low income. While it is not necessarily true that people with low income are not content with it, this study could be an indicator that the relation exists. This relation is not necessarily so important as well, but it is included in the model because it is expected to exist.

Null Hypothesis 17 A Dutch voter’s trust in political related content on Face-book has no relation with his or his or her cultural and social background Null Hypothesis 18 A Dutch voter’s trust in political related content on Twit-ter has no relation with his or his or her cultural and social background 2.3.3 Towards Social Trust

Moy & Scheufele (2000) report that watching television has a negative effect on social trust. On the other hand, reading newspapers and watching entertain-ment on television has a positive effect on social trust. Because social media is considered a new form of media, it is hypothesised that those also have an effect on social trust. Social trust shows a positive trust with correlation in political institutions and satisfaction with democracy (Zmerli & Newton, 2008). These relationships can argued to indeed exist, since less social trust can result in more polarisation within society, which on its turn in less trust in politics. Taking a look at the effects of social media on social trust would be an inter-esting side-effect of this study, since people can argue on social media, but also form communities and stay in touch.

Null Hypothesis 19 A Dutch voter’s social trust has no relation with his or her trust in political content on Facebook (H19a) or Twitter (H19b)

(16)

Null Hypothesis 20 A Dutch voter’s social trust has no relation with his or her trust in political content on news on television (H20a), current affairs programmes on television (H20b), talkshows on television (H20c), news on the radio (H20d), newspapers (H20e) or journals (H20f )

Null Hypothesis 21 A Dutch voter’s social trust has no relation with his or her cultural and social background

2.4 Measurement Model

Now that all indicators are defined and all hypotheses are stated, a measurement model can be proposed. The proposed model for this research can be found in figure 2.

(17)

3 Method

In order to validate the measurement model as presented in figure 2, a survey was conducted (N=131). All questions are based on the above described indicators, which should lead to a measure of the latent variable.

3.1 The Survey

The full survey is attached in Appendix A. The questions were all stated in Dutch, since the survey was conducted among Dutch citizens. The whole survey has been imported into Google Forms. An orange-coloured theme was used, since according to Robert Plutchik’s Wheel of Emotions, orange evokes feelings of interest and anticipation (Williams, 2013). This is just a small detail, but hopefully it increased participation.

This subsection first provides an outline on the structure of the questionnaire, followed by an explanation of what Likert scale is used and why, concluded by the sampling method.

3.1.1 Structure

The survey consisted out of five parts, based on the latent variables as described in figure 1. It started with a short introduction which stated a brief explanation on the aim of the research, followed by stating that a respondent should only continue if he or she was entitled to vote. Next to that, the participant was assured that participating in the survey was anonymous. An e-mail-address of the author was provided, in case a respondent would have questions or was interested in the outcome of the research.

The first section was based on the participant’s social and cultural back-ground, starting with asking for the year of birth and asking if the participant felt related to the Netherlands, an other Western country, a non-Western coun-try or just not related to the Netherlands at all. The questions that followed were all on Likert-scale base, regarding ethical diversity in the neighbourhood, contentment about living circumstances, income and social life. Then there was a question were the respondent had to put his neighbourhood on a scale of a working class neighbourhood to a villa district. Then participants had to rate themselves on their political orientation, on a scale from left to right. After this, the questions on social trust followed.

The second section was about trust in conventional media. For each medium (news on television, current affairs programmes on television, talkshows on tele-vision, news programmes on the radio, newspapers and journals) it was asked how often the respondent watches, listens or reads political-related news on that medium. This was a multiple choice question with the following options: every day, four to six days a week, one to three days a week, less often than above options and never. Then per medium the respondent had to rate on a Likert-scale how much he or she trusted the medium, to what extend tells the medium the whole story, how biased the medium is and the accuracy of the medium, as

(18)

described in section 2.2.3. For each dimension, a short explanation was given to the respondent, since the overlap between the dimension could have made it unclear what the differences were between them. It is important to note, that if a participant never listened to one of the media, he or she still had to fill out the questions regarding trustworthiness, completeness, bias and accuracy, based on a personal estimation.

Trust in social media is the third variable, which the third section of the questionnaire was about. The two social media that were taken in to account were Facebook and Twitter, as described in section 2.2.4. The questions to measure this trust were exactly the same as the questions regarding trust in conventional media.

The final part of the questionnaire was about trust in politics. Respondents were asked to rate eight political authorities (the current Dutch government, Dutch House of Representatives and Dutch Senate, political parties, politicians, local council, the Dutch legal system, the police and the European Union) on a Likert scale to what extend they trusted it. It was shortly described that if a respondent thought the concept of trust was too abstract, he or she should think in terms of: they want the best for you, they vouch for your safety and they do not have an hidden agenda. The final question was whether the participant voted in the Dutch national elections of March 2017. This was an interesting question, since the turnout for the whole population is known, namely 81,9% (Kiesraad, 2017).

3.1.2 Likert-scale questions

In the survey, some questions needed to be answered on a Likert scale, as de-scribed above. It is important to argue which scale was used, in particular whether is was a scale with a midpoint (5- or 7-point scale) or without one (4-, 6- or 8-point scale). Garland (1991) found that if you do not include a mid-point in the survey, the chance of receiving socially accepted answers from respondents, can be minimised. If a mid-point is not included, respondents are required to take a side, they cannot be neutral. He does stress that either including or omitting the mid-point in a survey, has a distorted effect on the results. One question in this research’s survey is whether the participant con-siders him- or herself politically left- or right-winged or neutral. By eliminating the neutral option here, one would have had to choose between being left or right winged, where neutral was a valid option. Next to that, the survey was conducted anonymously, with absence of the interviewer, which would reduce the urge to give a socially accepted answer to please the interviewer. Therefore, it was decided to include a mid-point in the survey.

Next, it had to be decided whether to use a 5-point or a 7-point scale. Both, and also other scales, are used in scientific work and there seems to be no standard. This does not mean that a random scale can be picked, as Krosnick & Presser (2010) found that some scale-lengths can maximise the reliability and validity of a survey (pp. 268). They go on arguing that it must be easy for a respondent to translate his or her opinion to the scale. To participants

(19)

with a very strong disagreement to the given statement, this is easy. However, people with a more moderate opinion, let’s say slightly agree, and with using a 5-point scale, they might rate a 4 on one question and 5 on the next in order to middle it out. By using a 7-point scale, people can either slightly or moderately (dis)agree.

Krosnick & Presser (2010) further report that reliability decreases growth on scales longer than 7-points. Also a 5- and 7-point scale are about equally reliable, as found by multiple authors. Regarding validity, they reported that longer-scales score better than smaller ones, but very long longer-scales may compromise validity. A 7-point scale, however, is not considered especially long. Therefore, and with take the other above arguments into account, it was decided to use a 7-point scale on the questions that needed to be answered on a Likert scale.

A 5-point scale, however, also has it pro’s. For instance, Babakus & Mangold (1992) write that a scale longer than 5-points may cause frustration among par-ticipants and that could cause decrease of the response-rate, since a respondent may quit the questionnaire. This is one example on the fact that there is not really consensus in scientific research which scale to use. For both scales there are arguments why they would be better or not. Anyway, the questionnaire of this research was not too long (about 50 scale questions, whereas this author is talking about 200 or longer), so this argument can be rejected.

3.1.3 Procedure

Respondents were collected using a convenient sampling method. The survey has been shared on the author’s social media channels (Facebook and What-sapp). This means that a part of the respondents were probably people from his social network and might be about same age. Therefore, it was important to stress that participating in the survey was anonymous, in order to rule out socially acceptable answers, as well as to reassure the respondent’s privacy. Be-cause the entire population of this sample is almost 13 million people (all Dutch people who are entitled to vote) (Kiesraad, 2017), it was desirable to have peo-ple from all ages from 18 and older. To achieve this, the peopeo-ple from different ages were targeted and asked if they would share the questionnaire with their friends and family. The author himself also sent the questionnaire to all of his former high-school teachers, who also vary from age. Finally, the survey has been shared in a Facebook group of the local city from which the author is from. The people in this group are connected by living in the same town, so this lead to responses over different ages, social backgrounds and level of education, which hopefully generated a more in depth and representative reflection of the population.

When a potential respondent received the link to the questionnaire, he or she could fill it in on a computer, smartphone or tablet. No supervision of an interviewer was necessary here. Once the respondent submitted the form, it automatically got registered by Google Forms. When the survey was closed, all submitted data has been downloaded in one comma-seperated (CSV) file. The survey has been online for 10 days.

(20)

3.2 Data processing

The downloaded CSV file, containing all answers, was downloaded and then pre-processed using Python. All column names were changed to shorter work-able strings, instead of whole questions, which made it difficult to work with. Next to that, some variables have been recoded and the age for every case has been calculated based on the year of birth. The recoded variables were: my neighbourhood is ethically varied (CSB3), political orientation (CSB6), people will profit from me when they get the chance (ST1) and most people only care for themselves (egoistic) (ST3).

4 Results

A total of 131 respondents filled out the questionnaire with a mean age of 37.6 years old. A first exploration of the data gives the age distribution, as seen in figure 3. The distribution is positively skewed, which can be explained because of the fact that a lot of the respondents are people from the author’s social network. Note that in this graph only the ages are included if at least one respondent had that age.

Figure 3: Age distribution

The ages are recoded into categories (>25, 26-35, 36-45, 46-55, 56-65 and 65>). Using One-Way ANOVA tests, no significant difference in trust in any medium has been found among the different age categories. Trust in per medium has been recoded into a new variable here, expressed as the mean of the trustwor-thiness, completeness, bias and accuracy scores regarding that medium. This

(21)

means every medium got a 1-7 total-trust score, in order to measure trust dif-ferences among the ages.

In the questionnaire it was asked, whether a respondent felt related to the Netherlands, an other Western or Non-western country or just not related to the Netherlands at all. As can be seen in figure 4, only 3.9% of the respon-dents felt more related to the an other Western country than the Netherlands, which corresponds with 4 people. One individual did not feel related to the Netherlands and neither to an other Western or non-Western country. All other respondents felt connected to the Netherlands.

Figure 4: Relation to NL

There is no significant difference in mean among people who did and did not vote and their trust in politics (p=.932) when equal variances are not assumed. The hypothesis that the variances are not equal can not be rejected, since the Levene’s test is not significant (p=.059). Note here that there were only 5 people in this sample who did not vote during the Dutch elections of March 2017. Respondents also had to report how frequent they used each medium. In figure 5 it is plotted what the mean of the four indicators was per medium grouped on frequency. Note that the amount of respondents per group is different, for example the group of CM6 that reads journals every day, consisted only out of one person. As can be derived from the plot, social media are considered less trustworthy than the conventional media.

Other than this exploration of the data, nothing has been done with the frequency of medium use, because it can be argued that people can deliber-ately not make use of a certain medium, because they do not trust it and their measures should be taken into account as well.

Although some interesting numbers were found on the frequencies of use of social media. 102 respondents never use Twitter as a news source and 35 people never use Facebook as a news source. Besides journals (which 76 respondents

(22)

never read) and news on the radio (to which 53 people never listen to), the other conventional media are still used more than social media in this sample.

Figure 5: Trust mean per medium per frequency of use

4.1 Measurements

The first step into working towards a valid model, is checking whether all mea-surements from the proposed measurement model in figure 2 fulfil the condi-tions. This subsection describes what those conditions are, as well as checking and deleting indicators and variables if necessary.

4.1.1 Reliability

Table 1 provides an overview of the internal reliability, convergent validity and the discriminant validity of the proposed measurement model. All values that do not meet the requirements as described in the following subsections, are in bold.

As showed in table 1, all latent variables can be considered reliable, since their alpha value is above .7 (Cronbach, 1947), except the variables social trust and cultural and social background.

For each indicator, the indicator reliability (IR), i.e. factor loading, has been calculated. This value should be greater than .709, because when this value is squared, the explained variance of that indicator to the variable will be greater than 0.5. For the CSB variable, all indicators, except for CSB4a, are below the desired value. These indicators have all been removed except for CSB4a and CSB4b (CSB4c’s loading factor was .592 < .709 when left in, so it is left out) and

(23)

the variables name is changed to Contentment (α=.651, CR=.826, AVE=.709, MSV=.219). Even though the alpha is just below .7, it is considered acceptable for next couple of iterations on the model. The IR of CSB4b is .701, which is slightly less than the .709 threshold. However, for now it is decided to leave it in. All other measures of contentment are fulfilling the conditions. The remaining CSB indicators have been removed from the model.

For the ST variable, ST1 and ST3 do no meet the requirement of being greater than .709, so these are removed, which resulted in slight alpha increase for social trust (α=.625, CR=.840, AVE=.725, MSV=.288). Indicators TP5 and TP7 have been removed from this model as well, which affected the trust in politics variable (α=.910, CR=.931, AVE=.693)

4.1.2 Validity

According to Hair et al. (2010), the composite reliability (CR) should be greater than .7. This accounts for all variables, except for CSB, which is already removed from the model now. The new defined contentment variable however does meet the requirements. The threshold value of the average variance extracted (AVE) is .5 to be considered valid (Bagozzi & Yi, 1988). This condition is not met by the deleted CSB variable, as well as ST. However, the requirement is met for ST after the revision of this variable as described in the previous section.

The AVE should be greater than both the average squared variance (ASV) and the maximum square variance (MSV). However, the ASV itself can not be higher than the MSV and because no MSV is higher than any AVE, ASV is not included in table 1.

4.2 SEM Analysis

Now that all measures are considered sufficient, the model with its adjusted variables can be run. Table 2 shows the hypotheses as stated in the related work section, with the corresponding path coefficients (β) and probability values (p). The path coefficient of an hypothesis has only been displayed if the relation was significant. The explained variance of trust in politics in this model is 45,7%. All significant relations in this model have an effect size (f2_{) between .02 and}

.15, which is considered a small effect (Cohen, 1992).

The inner VIF-values are also reported in table 2. Those values should be smaller than 5, otherwise they indicate collinearity among the indicators (Sarstedt et al., 2017). As can be seen in the table 2, all variables fulfil this requirement.

Note the hypotheses regarding the removed variable CSB have not been included in table 2. Even though two of its indicators are still being used for the contentment variable, that variable is considered substantially different than CSB. Therefore the hypotheses regarding contentment are undefined, since they have not been defined in the related work section of this paper. All null hypotheses regarding the CSB variable (H4, H6, H8, H10, H12, H14, H16, H17, H18, H21) are in this stage of the research failed to be rejected.

(24)

Table 1: Calculated Cronbach’s alpha, composite reliability, average variance ex-tracted & maximum square variance per latent variable and indicator reliability per indicator for the proposed model

Latent Variable Indicator α IR CR AVE MSV Cultural and Social Background CSB3 .516 .232 .625 .240 .334

CSB4a .719 CSB4b .458 CSB4c .320 CSB5 .395 CSB6 .631 Social trust ST1 .622 .479 .770 .466 .297 ST2 .799 ST3 .594 ST4 .802

Trust in TV News CM1a .897 .880 .929 .765 .687

CM1b .880

CM1c .836

CM1d .900

Trust in TV Current Affairs CM2a .929 .992 .949 .823 .540

CM2b .917

CM2c .889

CM2d .903

Trust in TV Talkshow CM3a .907 .910 .934 .779 .650

CM3b .894

CM3c .811

CM3d .912

Trust in Radio News CM4a .959 .958 .970 .890 .693

CM4b .931

CM4c .949

CM4d .937

Trust in Newspapers CM5a .935 .922 .953 .836 .540

CM6b .928

CM6c .902

CM6d .905

Trust in Journals CM6a .916 .936 .942 .803 .486

CM6b .948

CM6c .773

CM6d .916

Trust in Facebook SM1a .941 .917 .958 .849 .614

SM1b .928

SM1c .913

SM1d .928

Trust in Twitter SM2a .949 .931 .963 .868 .176

SM2b .943 SM2c .942 SM2d .909 Trust in politics TP1 .898 .877 .920 .595 TP2 .893 TP3 .838 TP4 .824 TP5 .609 TP6 .771

(25)

Table 2: SEM analysis results, hypotheses with their path coefficients, effect sizes, inner VIF values and p-values

Hypothesis β f2 VIF p

H1a trust in Facebook → trust in politics .021 1.888 .121 H1b trust in Twitter → trust in politics .003 1.842 .503 H2a trust in TV news → trust in politics .002 2.616 .671 H2b trust in TV current affairs → trust in politics .213 .028 3.021 .060 H2c trust in TV talkshow → trust in politics .009 1.925 .308 H2d trust in radio news → trust in politics .014 2.601 .215 H2e trust in newspaper → trust in politics .024 2.337 .139 H2f trust in journal → trust in politics .021 1.803 .146 H3 social trust → trust in politics .160 .040 1.189 .097 H5a trust in Facebook → trust in TV news .180 .022 1.615 .065 H5b trust in Twitter → trust in TV news .007 1.614 .291 H7a trust in Facebook → trust in TV current affairs .301 .065 1.615 .022 H7b trust in Twitter → trust in TV current affairs .004 1.614 .498 H9a trust in Facebook → trust in TV talkshow .028 1.615 .137 H9b trust in Twitter → trust in TV talkshow .017 1.614 .207 H11a trust in Facebook → trust in radio news .017 1.615 .109 H11b trust in Twitter → trust in radio news .217 .034 1.614 .020 H13a trust in Facebook → trust in newspaper .300 .068 1.615 .003 H13b trust in Twitter → trust in newspaper .014 1.614 .217 H15a trust in Facebook → trust in journals .000 1.615 .849 H15b trust in Twitter → trust in journals .308 .066 1.614 .008 H19a trust in Facebook → social trust .028 1.837 .106 H19b trust in Twitter → social trust -.275 .051 1.752 .035 H20a trust in TV news → social trust .000 2.616 .905 H20b trust in TV current affairs → social trust .008 2.996 .380 H20c trust in TV talkshow → social trust .008 1.910 .442 H20d trust in radio news → social trust .318 .049 2.480 .051 H20e trust in newspaper → social trust .002 2.332 .619 H20f trust in journals → social trust .002 1.800 .712 undef contentment → trust in politics .005 1.135 .434 undef contentment → trust in TV news .179 .034 1.029 .035 undef contentment → trust in TV current affairs .173 .033 1.029 .055 undef contentment → trust in TV talkshow .017 1.029 .130 undef contentment → trust in radio news .213 .051 1.029 .020 undef contentment → trust in newspaper .193 .054 1.029 .007 undef contentment → trust in journals .167 .031 1.029 .046 undef contentment → trust in Facebook -.151 .023 1.000 .091 undef contentment → trust in Twitter .023 1.000 .130 undef contentment → social trust .167 .030 1.102 .078

4.2.1 Model Adjustments

Based on the first analysis of the proposed model, all insignificant relations (p>.10) have been removed. This resulted in the removal of the latent variable trust in TV talkshows as a whole, because this variable turned out to have

(26)

no incoming or outgoing significant relation. Based on this adjustment of the model, null hypotheses H1a, H1b, H2a, H2c, H2d, H2e, H2f, H5b, H7b, H9a, H9b, H11a, H13b, H15a, H19a, H20a, H20b, H20c, H20e and H20f could not be rejected.

After this adjusment, the SEM analysis has been runned again. The ex-plained variance of trust in politics dropped to 34.9%, which can be exex-plained because in this new model it only has two relations incoming. In this new model, the relationship from contentment to trust in Facebook (p=.101) and social trust (p=.113) turned out to not be significant, so they have been deleted. The same goes for trust in Twitter to social trust (p=.137). This means that H19b can not be rejected.

In the next iteration, contentment to trust in TV current affairs was insignif-icant (p=.125), so this relation has been removed for the next iteration. This lead to a model of which the results can be seen in table 3. So far, relations with p<.10 were considered significant. However, the new composed variable contentment does turn out to have no effect on the explained variance of the dependent variable trust in politics. The relation from contentment to trust in newspapers also exists, but the effect is considered small (f2_{=.052). Since}

the alpha of the contentment variable is doubtful, the variable does not show a significant relation to the dependent variable and no hypotheses are formed re-garding this hypothesis, contentment is considered irrelevant for the final model of this research. Based on this argumentation, the variable contentment is not included in the final model of this research. This slightly affected the path co-efficients, effect sizes and inner collinearity values of the media variables in the final model.

H2b trust in TV current affairs → trust in politics .524 .410 1.028 .000 H3 social trust → trust in politics .198 .059 1.028 .033 H5a trust in Facebook → trust in TV news .246 .064 1.023 .002 H7a trust in Facebook → trust in TV current affairs .321 .115 1.000 .001 H11b trust in Twitter → trust in radio news .316 .112 1.024 .000 H13a trust in Facebook → trust in newspapers .385 .175 1.023 .000 H15b trust in Twitter → trust in journals .328 .120 1.024 .000 H20d trust in radio news → social trust .278 .084 1.000 .000 undef contentment → trust in TV news .179 .034 1.023 .076 undef contentment → trust in radio news .216 .052 1.024 .090 undef contentment → trust in newspapers .206 .050 1.023 .029 undef contentment → trust in journals .172 .033 1.024 .056

Table 3: SEM (including contentment) analysis results, hypotheses with their path coefficients, effect sizes, inner VIF values and p-values

4.3 Final Model

The final model which is worked towards in this research is shown in figure 6. As can be seen in table 4, all relations within this model are significant (p<.05).

(27)

All inner VIF values also meet the requirements of being smaller than 5 and no effect size is smaller than .02, what would indicate that there is no effect.

As can be seen in table 4, trust in TV current affairs shows a positive relation to trust in politics (p=.000), with a path coefficient of .524 and an effect size of .410 (>.350), which is considered a large effect (Cohen, 1992). Therefore hypotheses H2b is rejected. Social trust also shows a significant relation to trust in politics (p=.034). Though the effect of that relation is considered small (f2_{=.059, β=.198), hypothesis H3 is rejected as well.}

Trust in Facebook to trust in newspapers is significant (p=.000) and has a medium effect (f2_{=.175). The path coefficient of this relation is .385, which can}

be considered moderate. Based on these data, H13a is also rejected.

The remaining relations all show a small effect (f2 _{values between .02 and}

.15), which means there is an effect (Cohen, 1992) and therefore the null hy-potheses H5a, H7a, H11b, H15b and H20d are rejected, even though the effects are small.

Trust in politics’s coefficient of determination (R2_{) in this model is 34.8%. In}

table 5 all the coefficients of determination per endogenous construct have been listed, as well as the predicted relevance (Q2_{) for each of these variables. All}

Q2are greater than zero, which means the exogenous constructs have predictive relevance for that concerned endogenous variable (Hair et al., 2011).

H2b trust in TV current affairs → trust in politics .524 .410 1.028 .000 H3 social trust → trust in politics .198 .059 1.028 .034 H5a trust in Facebook → trust in TV news .220 .051 1.000 .006 H7a trust in Facebook → trust in TV current affairs .321 .115 1.000 .001 H11b trust in Twitter → trust in radio news .284 .087 1.000 .000 H13a trust in Facebook → trust in newspapers .357 .146 1.000 .000 H15b trust in Twitter → trust in journals .304 .102 1.000 .000 H20d trust in radio news → social trust .278 .084 1.000 .000

Table 4: Final model SEM analysis results, hypotheses with their path coefficients, effect sizes, inner VIF values and p-values

Variable R2 Q2 Trust in TV news .048 .029 Trust in TV current affairs .103 .077 Trust in radio news .080 .063 Trust in newspapers .127 .095 Trust in journals .092 .066 Social trust .077 .041 Trust in politics .348 .218

Table 5: Coefficients of determination and predicted relevance per endogenous vari-able of the final model

(28)

Figure 6: Final Model

When the contentment variable would be still in the model, the coefficients of determination and predicted relevance values of trust in TV news, radio news, newspapers and journals would be slightly higher, but the variables are still greater than zero.

5 Discussion

In this research it is probed to find out to what extend the use of social media relative to conventional media as a news source affected the trust of the Dutch public in its national politics. This research failed to reject the null hypotheses H1a and H1b, respectively the two hypotheses that indicated the relation be-tween trust in Facebook and Twitter to trust in politics. The null hypotheses

(29)

regarding trust in conventional media to trust in politics (H2a, H2c, H2d, H2e and H2f) were also failed to be rejected. H2b, regarding trust in TV current affairs does not relate to trust in politics, however, was rejected. All together it can be concluded that for now a significant relation between trust in social media as a news source for political content and trust in Dutch politics could not be detected. The hypothesis that social trust does not have a relation with trust in politics, however, was rejected, though this is not a form of media.

It was also tested whether Facebook or Twitter had an effect on the con-ventional media types. It turned out that the hypotheses regarding trust in Facebook does not have a relation to trust in TV news, current affairs pro-grammes on TV and newspapers could be rejected. The same accounts for the relation between trust in Twitter and trust in radio news and journals. Lastly, the hypothesis that the relation between trust in radio news towards social trust does not exist has been rejected as well.

In the remaining of this paper the limitations of this research are discusses, as well as recommendations for future research.

Some respondents provided direct feedback on the questionnaire to the au-thor, since some of the participants are from the author’s social network. The main remark was that respondents did not understand why they had to fill out the trustworthiness, completeness, bias and accuracy of a medium they never use. Even though it was stated in the questionnaire itself that they had to give an estimation, some filled out the midpoint (4) on all dimensions for every medium they did not use. This may have caused some noise in the results. This would also be an argument to not include the midpoint.

Of all different conventional media variables, the journal scored the lowest mean on its indicators (3.8 on a 1 to 7 scale), after talkshows on TV, while in general, journals could be considered rather reliable. In Dutch a journal is literally translated called an opinion magazine. Even though some examples of Dutch journals were provided in the questionnaire, it might be the case that some respondents thought this medium regarded tabloids. It is notable that the only person who reads journals on daily base did rate journals high.

Other than that, a lot of the used literature on trust in politics was about a partisan choice for example in the United States, while in the Netherlands there were 28 parties participating in the elections of March 2017. With this amount of parties, the difference between for instance left and right might be too simplistic. It might have been better to also make a difference between progressive and conservative as well as between liberalism and statism.

Regarding the questionnaire, it would have been interesting to include a question regarding a respondent’s highest level of educational degree as a part of the CSB variable and see if there were any differences between those groups in how much they trusted conventional and social media and politics. Furthermore, if this research would be repeated, more in depth research is required on the CSB variable, that had to be deleted from the results of this research. However, it is still arguable that a voter’s cultural and social background influences his

(30)

or her trust in politics. This point actually already concerns a recommendation for future work in this field.

As described in the related work section, Youtube is not taken into account in this research. In hindsight however, it might have been good to investigate it. The content on Youtube could be uploaded by anyone. Next to that, the content that is recommended to its users is based on the filter bubble of that user. There is political content available on Youtube and the comment section on each video offers a platform for interpersonal communication. The political content on Youtube can vary from debates to conspiracy theories, so these could have an effect on trust in politics.

The variable contentment is removed from the final model, however, if this research would be repeated with a larger sample, it is recommended to included the variable in again, since it is likely that the probability value will decrease, whereas it was greater than .05 in this research. The social trust variable re-mained debatable after the revisions made in this research with respect to its usage in different researches. However, the ’regular’ way of measuring social trust was not validated. It can be concluded that this way of measuring social trust is also not ideal.

This research focused on perceived trust, i.e. how do people evaluate their own trust in the different media types. Those results could be different than the actual effects in reality. Respondents could have judged social media less reliable than conventional media, because they felt this is the way it is perceived in general, or the way it should be, while unconsciously it could affect their behaviour or opinion.

5.1 Recommendations

For future research it is recommended to split out the different social media into more latent variables than only Facebook and Twitter. A difference can be made between content posted by people and companies. Companies on their turn can be either verified or not. Also most newspapers nowadays post content on social media. By those developments the borders between traditional and new media become unclear. It would be good to investigate if trust in content differs for printed newspapers, posted on the newspaper’s website and their social media. As described in the related work section, some conventional media channels start using social media content as news. It comes down to all news sources getting intertwined with each other. Developing scientific guidelines on the different social media types is desirable.

Since the usage of the different types of social media differs per country, it would be interesting to conduct this research in different countries, which could be compared to each other. The conventional media channels and programmes are different per country, as well could be the presence of politicians on social media. The Dutch prime minister for example uses Twitter, but his usage is not as prominent and controversial as Donald Trump’s Twitter usage.

Lastly, it is recommended to research the effects of social media on trust in politics or voting behaviour in general. Post-truth politics, as described in

The Influence of Social Media Usage as a News Source on Trust in Politics