Social Media, The Real Enemy? “What mechanisms explain the dissemination and correction of rumours on Twitter during the tram shooting crisis in Utrecht?”

Social Media, The Real Enemy?

“What mechanisms explain the dissemination and correction of rumours on Twitter during the tram shooting crisis in Utrecht?”

Name: Anouk van Twist Student number: S2456729

MSc course: Thesis Crisis and Security Management Thesis supervisor: Stef Wittendorp

Second reader: Dr. Sanneke Kuipers Submission date: January 7th_{, 2020}

Table of Content

2.1 Social Media and Crisis Communication ... 12

2.1.1. A Definition of Crisis ... 12

2.1.2. A Definition of Crisis Communication ... 13

2.1.3. A Definition of Social Media ... 13

2.1.4. Social Media and Crisis Communication ... 14

2.2 Functions of Social Media during Crisis ... 14

2.3 Information Disorder: mis-, dis-, mal- information ... 16

2.4 The Self-Correcting Mechanisms Thesis ... 19

2.5. The Official Dominance Thesis ... 20

2.6. Research Expectations ... 21

3. Methodology ... 23

3.1 Research Strategy ... 23

3.2 Research Design: A Single Case Study ... 23

3.3 Case Selection ... 24

3.4 Within-Case Analysis: mis-, dis-, and mal- information ... 24

3.5 Social Media Platform Twitter ... 25

3.6 Self-Correcting Mechanism: Operationalisation ... 26

3.7 Official Dominance: Operationalisation ... 27

3.8 Data Collection ... 29

3.9 Data Analysis ... 30

3

3.11 Research Limitations ... 31

4. Social Media Analysis: The Utrecht Tram Shooting ... 34

4.1 The Utrecht Tram Shooting ... 34

4.2 Social Media Use During the Utrecht Tram Shooting ... 35

5. Misinformation ... 37

5.1 Misinformation 1: There are Multiple Shooting Sites ... 37

5.2 Misinformation 2: There is a Shooting at a Mosque ... 42

5.3 Conclusion Misinformation ... 45

6. Disinformation ... 47

6.1 Disinformation 1: Suspect in Utrecht Published Manifesto Online ... 47

6.2 Disinformation 2: Militant Dutch White Nationalist, Sam Hyde, Claims Responsibility ... 52

6.3 Conclusion Disinformation ... 56

7. Malinformation ... 57

7.1 Malinformation 1: Turkish and Moroccan Boys Celebrate the Tram Shooting ... 57

7.2 Malinformation 2: The Tweet by Fake Account @PolitieP: All Mosques Closed ... 62

7.3 Conclusion Malinformation ... 66

8. Discussion ... 67

8.1 Overview ... 67

8.2 Social Media, The Real Enemy? ... 70

9. Conclusion ... 72

9.1 Answer to Research Question ... 72

9.2 Research Limitations ... 73

9.3 Academic Relevance and Future Research ... 75

9.4 Practical Recommendations ... 76

References ... 79

Appendix ... 86

Annex 1: Codebook ... 86

Table of Figures

Figure 1: The Classification of Information Disorder (Wardle and Derakshan, 2017). ... 18

Figure 2: Timeframe of Misinformation 1 (There are Multiple Shooting Sites) Dissemination on Twitter in the Utrecht Tram Shooting Crisis. ... 38

Figure 3: The Number of Tweets that Confirmed or Disputed Misinformation 1 (There are Multiple Shooting Sites) Throughout the Crisis. ... 40

Figure 4: Timeframe of Misinformation 2 (There is a Shooting at a Mosque) Dissemination on Twitter in the Utrecht Tram Shooting Crisis. ... 43

Figure 5: The Number of Tweets that Confirmed or Disputed Misinformation 2 (There is a Shooting at a Mosque) Throughout the Crisis. ... 44

Figure 6: Timeframe of Disinformation 1: (Suspect in Utrecht Published Manifesto Online) Dissemination on Twitter in the Utrecht Tram Shooting Crisis. ... 48

Figure 7: Timeframe of Disinformation 2: (Militant Dutch White Nationalist, Sam Hyde, Claims Responsibility) Dissemination on Twitter in the Utrecht Tram Shooting Crisis. ... 53

Figure 8: Timeframe of Malinformation 1 (Turkish and Moroccan Boys Celebrate the Tram Shooting) Dissemination on Twitter in the Utrecht Tram Shooting Crisis. ... 58

Figure 9: The Number of Tweets that Confirmed or Disputed Malinformation 1 (Turkish and Moroccan Boys Celebrate the Tram Shooting) Throughout the Crisis. ... 60

Figure 10: Timeframe of Malinformation 2 (The Tweet by Fake Account @PolitieP: All Mosques Closed) Dissemination on Twitter in the Utrecht Tram Shooting Crisis... 63

Figure 11: The Number of Tweets that Confirmed or Disputed Malinformation 2 (The Tweet by Fake Account @PolitieP: All Mosques Closed) Throughout the Crisis. ... 64

Figure 12: Functions of social media during crisis (Houston et al, 2015) ... 91

Table 1: Selected Rumours Utrecht Tram Shooting. ... 25

Table 2: The Operationalisation of the Self-Correcting Mechanisms Thesis for the Utrecht Tram Shooting. ... 27

Table 3: The Operationalisation of the Official Dominance Thesis for the Utrecht Tram Shooting. ... 29

Table 4: Collected Tweets per Rumour. ... 30

Table 5: Twitter Accounts That were Most Apparent during the Utrecht Tram Shooting. ... 35

Table 6: Overview Misinformation 1: There are Multiple Shooting Sites. ... 41

Table 7: Overview Misinformation 2: A Shooting at a Mosque. ... 45

Table 8: Overview Disinformation 1: Suspect in Utrecht Published Manifesto Online. ... 51

Table 9: Overview Disinformation 2: Militant Dutch White Nationalist, Sam Hyde, Claims Responsibility. ... 55

Table 10: Overview Malinformation 1: Turkish and Moroccan Boys Celebrate the Tram Shooting. .. 61

Table 11: Overview Malinformation 2: The Tweet by Fake Account @PolitieP: All Mosques Closed. ... 65

Table 12: Overview of the Analysed Rumours in Relation to the Theory. ... 68

Photo 1: Picture that factchecker @JPeterBurger added to his tweet... 48

Photo 2: Tweet by @HavenMoahan with photo of alleged perpetrator Sam Hyde ... 52

5

Abstract

Social media have been recognised as important crisis communication channels. On the one hand, social media can be used to share information, advice, news items and personal experiences during a crisis situation. But, on the other hand, social media can also be used to disseminate mis-, dis-, and mal- information during crisis. Consequently, information disorder on social media could further disrupt the crisis situation. Crisis communication professionals need to understand how rumours are disseminated on social media platforms during crisis. Despite this obvious need, there is scarce research available on the dissemination of rumours on social media and the possibility of social media communities and authorities to correct mis-, dis-mis-, and mal- information. Thusmis-, to answer this needmis-, this exploratory research examines six rumours that circulated on Twitter during the Utrecht tram shooting. The findings suggest that mis-, dis-, and mal- information was corrected on social media during the Utrecht tram shooting. Remarkably, government authorities were mostly absent in the correction of rumours on social media.

Keywords: crisis communication, Twitter, social media, the Netherlands, rumour, self-correcting mechanisms thesis, official dominance thesis.

Acknowledgements

From September 2019 till December 2019 I worked on the research report: “Social Media, The Real Enemy?”. This research report is the final completion of the master ‘Crisis and Security Management’ at Leiden University. In this thesis, I analysed how rumours were disseminated and corrected on Twitter during the Utrecht tram shooting.

Writing this research report was an interesting experience for multiple reasons. First, I had the opportunity to investigate a topic that I find fascinating, namely the role of authorities in the fight against fake news. Second, writing this research report was an educational experience because I have learned to conduct research independently and apply acquired disciplinary knowledge and methodological skills.

Completing this research would not have been possible without the support of my supervisors. Therefore, I would like to thank Stef Wittendorp and my second reader Sanneke Kuipers. They guided me through the research process and supported me with feedback and advice. Besides, I would like to thank my family for their positive support during the whole process.

I hope you enjoy reading this thesis. With kind regards,

7

1. Introduction

On March 18th_{, 2019, a man started shooting in a tram in Utrecht. Consequently, three people}

were killed, and seven others were injured. The man hijacked a car and fled the shooting site. Immediately, a major police operation was initiated to find the perpetrator. The police indicated that the shooting could be a terrorist attack. As a result, the terrorism threat level in Utrecht was increased and people in Utrecht were advised to stay inside. Schools, shops and mosques were closed. Around 18:45, Gökmen Tanis, a 37-year-old man, was arrested. During the crisis, social media were used by government officials, media authorities and citizens to disseminate crisis information, share practical advice, and to find the perpetrator. However, meanwhile, also false information was disseminated on social media. For example, rumours were spread about several shooting sites, about a white nationalist, Sam Hyde, who had claimed responsibility, and about Turkish and Moroccan people celebrating the attack.

The Utrecht tram shooting is an example of a crisis in which social media played a role in the dissemination of both true and false information. This phenomenon is present during many different crisis situations.

A striking example is the terrorist attack on the 11th _{of September 2001. That day, two planes} flew into the World Trade Centre (WTC), a third plane hit the Pentagon and a fourth plane crashed into a field. During the crisis, social networking platforms, predominantly via the internet, were used by government authorities and citizens to share information about the event, and to gather information about missing people (Reuter, Hughes & Kaufhold, 2018). However, during the crisis, also rumours and conspiracies were shared on social media platforms (Emery, 2017). For instance, conspiracy thinkers shared that the US government deliberately allowed the hijacked planes to reach their targets (Olmsted, 2019). Other rumours concerned unbelievable stories of people who survived the fall of the WTC. For example, there were rumours about a policeman who had reportedly fallen from the 82nd _{floor and had only broken} his legs and another rumour about a man who surfed down from the 50th _{floor (Emery, 2017).}

Another instance is the fire at the Notre Dame in Paris in 2019. During the crisis, a lot of people used Twitter to express their support for the French people and social media were used for fundraising money. However, there were also tweets spreading rumours and conspiracies. For instance, in many tweets, Muslims were said to be the villains that started the fire. There were

even videos posted in which a Muslim was present in the cathedral suggesting that he lit the fire. In addition, there were people who claimed that the Notre Dame was deliberately lit to divert attention from the fire in the Al-Aqsa mosque (holy place in Jerusalem). The conspiracy behind this suggested that Israeli Prime Minister Netanyahu would have had plans to replace the mosque with a (Jewish) temple (NOS, 2019).

Lastly, an indicative example of a crisis situation in which true and false information was disseminated on social media is the gunman in the newsroom crisis in the Netherlands. In 2015, a man with a gun entered a tv studio in the Netherlands. The man demanded airtime during the evening news. The TV channel decided not to broadcast what was happening. Consequently, at the same time, on social media platforms the crisis in the television studio was intensively discussed. On the one hand, the discussion online was helpful because the spread of relevant information, official statements and debates about the possible perpetrator were started. However, on the other hand, two rumours were spread intensively on Twitter during the crisis: firstly that the parents of the gunman were killed in the MH17 tragedy, and secondly that an identical situation had taken place in Belgium (Jong & Dückers, 2016). Both rumours appeared not to be true in hindsight. Jong and Dückers (2016) concluded that social media platforms in this case did not only stimulate the dissemination of rumours, but also revealed that social media communities were able to correct false information during crisis. However, interestingly, Jong and Dückers (2016) did not investigate the role of authorities in the correction of the rumours. According to Korthagen (2015) authorities and their messages, including those correcting rumours, may dominate social media during crisis because authorities are trusted and legitimate sources of crisis information.

1.1 Problem Definition

Scholars and government officials have recognized the importance of social media for crisis communication (Takahashi, Tandoc & Carmichael, 2015; Beneito-Montagut, Anson, Shaw & Brewster, 2013). During a crisis, social media can be used by government officials and citizens in several ways. According to Alexander (2014), Bruns, Burgess, Crawford & Shaw (2012) and Houston et al. (2015) social media could be used positively during crises to listen to public debate, monitor the situation and create cohesion. However, social media could also pose a threat, because mis-, dis-, and mal- information can spread easily on these platforms (Alexander, 2014; Prooijen & Douglas, 2017). Rumours are characterised as misinformation,

9 example of misinformation is a message that is misunderstood or misinterpreted (Wardle & Derakhshan, 2017). Disinformation occurs when false information is deliberately disseminated online with a harmful intent. Specifically, disinformation often includes messages that use a false context, imposter content, and manipulate or fabricate content (Sellnow, Parrish & Semenas, 2019). Information is characterised as malinformation, when unverified or genuine information is shared with the intent to do harm; take for example conspiracy theories, harassments and hate speech (Wardle & Derakhshan, 2017).

Recently, Jong and Dückers (2016) have argued, based on a single case study, that social media platforms might also be able to debunk and correct rumours. As a result, social media might both stimulate and correct the dissemination of false information during crisis situations. However, there is little research available to understand the dissemination of rumours on social media and the possibility of social media communities to correct mis-, dis-, and mal- information during crisis. As of such, this exploratory research will examine the dissemination and the correction of information on social media during the Utrecht tram shooting and the role of authorities therein.

1.2 Research Question

I would like to gain insight into how rumours are disseminated and corrected on social media platforms during crises. To understand this, I would also like to analyse which actors are actively involved in the correction of rumours and whether government authorities on social media try to influence the dissemination of false crisis information online. Therefore, I would like to conduct a social media analysis in relation to the Utrecht tram shooting. The following research question will be studied: “What mechanisms explain the dissemination and correction of rumours on Twitter during the tram shooting crisis in Utrecht?”

To answer the main question, the following sub-questions will be examined:

(1) How are rumours disseminated on Twitter during the tram shooting crisis in Utrecht? (2) How are rumours corrected on social media during the tram shooting in Utrecht?

(self-correcting mechanisms thesis)

(3) Which actors were involved in the correction of rumours on social media during the tram shooting in Utrecht? (official dominance thesis)

1.3 Academic Relevance

Little is known about the functions of social media during crises (Binder, 2012). Alexander (2014), Bruns et al. (2012) and Houston et al. (2015) argue that social media can be used by government authorities and citizens in several positive ways. However, social media also have disadvantages, because mis-, dis-, and mal- information can be disseminated easily on these platforms during crises (Alexander, 2014; Prooijen & Douglas, 2017). This thesis will analyse the self-correcting mechanisms thesis and the official dominance thesis to understand which mechanisms could explain the dissemination and the correction of rumours on social media First, the self-correcting mechanisms thesis argues that social media communities are able to correct rumours on social media during a crisis situation. Jong and Dückers (2016) have argued, based on a single-case study, that online communities can correct false information via three mechanisms: the sender of the incorrect tweet can try to correct the rumour, other Twitter users can try to validate the information and can correct rumours, and users on social media can help further disseminate or challenge the corrected messages on social media during crisis. More academic research into the topic of rumours and correcting mechanisms of social media platforms during crises may help to confirm these conclusions and improve current understanding about the challenges that social media pose during crisis. Second, the official dominance thesis argues that (correction) messages of authorities dominate traditional media during crisis because their information is trusted and therefore their messages are often further disseminated (Korthagen, 2015). However, little is known about the possibilities of the official dominance thesis to explain the correction of crisis information on social media as well.

1.4 Societal Relevance

In the event of a crises, social media may be used to disseminate mis-, dis-, and mal- information (e.g. Alexander, 2014; Bruns et al., 2002; Bird et al., 2012; Starbird et al., 2014;). This could worsen a crisis situation and could have devastating consequences for crisis management. Therefore, this research report could advise authorities on their possible capabilities to correct and/or hinder the dissemination of rumours during crisis (Sunstein, 2009). Moreover, this research can inform citizens about the pros and cons of social media during crises and provide insights into the possibilities for Twitter users to correct rumours.

11 1.5 Reading Guide

This research report consists of nine chapters. The research theme and the relevance of the research question were explained in this introduction chapter. The next chapter will examine relevant literature. Subsequently, the third chapter will focus on the research strategy, research method, data collection, data analysis, research ethics and research limitations. The findings from the social media analysis will then be presented in the results chapters. Thereafter, remarkable findings will be discussed in the discussion chapter and the research question will be answered in the conclusion. This research report will end with some recommendations for practice and possible follow-up research.

2. Theoretical Framework

To examine the dissemination of rumours on social media during crisis this chapter will provide an overview of relevant literature, define relevant concepts and discuss how the concepts may theoretically relate to each other. First of all, the concepts ‘crisis’, ‘crisis communication’ and ‘social media’ will be defined. Second, the use of Twitter (as a specific platform of social media) during crisis will be analysed and opportunities and threats are identified. Third, the concept of information disorder is described (more specifically: mis-, dis-, mal- information). Thereafter, the self-correcting mechanisms thesis and the official dominance thesis will be explained to understand which mechanisms might help explain the correction of rumours on social media. This chapter will conclude with some research expectations.

2.1 Social Media and Crisis Communication 2.1.1. A Definition of Crisis

Crisis has shown to be hard to define as there is no universally agreed upon definition of the concept ‘crisis’. This could be the case because there are a variety of types of crises that entail different aspects. Additionally, crises can be perceived differently depending on the actor, resulting in largely divergent conceptualisations. For instance, Coombs (2007) focuses on the organisational perspective of crisis. He defines crisis as the “the perception of an unpredictable event that threatens important expectancies of stakeholders and can seriously impact an organization’s performance and generate negative outcomes” (Coombs, 2007, p. 2–3). Also, Fearn-Banks (2016, p.1) defines crisis as “a major occurrence with a potentially negative outcome affecting an organization, company, or industry, as well as publics, products, services or good name. It interrupts normal business transactions and can sometimes threaten the existence of the organization”. Both scholars are primary focused on organisational logics, such as maintaining existence and legitimacy during crisis. Boin (2005) defines crisis from a governmental perspective. He defines crisis as “a serious threat to the basic structures of the fundamental values and norms of a system, which under time pressure and highly uncertain circumstances necessitates making vital decisions” (Boin, 2005, p. 2). Thus, according to Boin (2005) an event can be classified as a crisis if it evokes feelings of threat, uncertainty and urgency by both government authorities and citizens. The definition of Boin (2005) will be used in this research report because it focuses on the broader (non-organizational) perspective of crises.

13 2.1.2. A Definition of Crisis Communication

Crisis communication is “the collection, processing, and dissemination of information required to address a crisis situation” (Coombs & Holladay, 2010, p.20). Crisis communication can be used crisis, during crisis and post-crisis. According to Coombs and Holladay (2010), pre-crisis communication involves collecting information and training people about risks. Crisis communication during a crisis involves the collection of information for crisis teams and the dissemination of information to citizens. Post-crisis communication involves follow-up crisis messages, sense making and meaning making (Coombs & Holladay, 2010). For the purposes of this research, I will define crisis communication as the collection, processing and dissemination of information required to address a crisis situation shared by government authorities and citizens. Thereby, I will focus on crisis communication during the crisis situation. Pre-crisis communication and post-crisis communication will not be examined because the objective of this study is to understand the dissemination of rumours during crisis. With the help of a social media analysis only crisis communication on social media during the crisis is analysed to understand the impact of rumours on crisis situations.

In the past, crisis communication was disseminated through traditional communication channels. Government authorities disseminated crisis information through television, radio and newspapers. Crisis information was disseminated by an official sender (government/media authority) to the receivers (citizens) without the possibility of direct feedback from receivers. As a result, in the past, authorities were the only producers of crisis information (Sellnow & Seeger, 2013). Social media have changed crisis communication profoundly. Today, citizens are not only receivers of crisis information. Social media enable citizens to actively engage in the production of crisis information (Lewandowsky, Ecker, Seifert, Schwarz & Cook, 2012; Sellnow & Seeger, 2013). As a result, social media users are able to seek, share and generate information during crises. Nowadays this creates an extra challenge for crisis communication because the government is not the only actor that disseminates information during crisis. Crisis communication is now interactive, networked and can follow a two-way dialogue.

2.1.3. A Definition of Social Media

Social media is “an umbrella term that is used to refer to a new era of Web-enabled applications that are built around user-generated or user-manipulated content, such as wikis, blogs, podcasts, and social networking sites” (Jin, Liu & Austin, 2014, p.75). In line with this, Alexander (2014) argues that social media are characterised by interactive communication

because social media allow messages to be shared and exchanged between individuals and organisations. As a consequence, social media users are simultaneously information seekers and information generators. On the one hand, users can consume information on social media that other users disseminated among the social media community. On the other hand, users are also able to generate information by sharing information, ideas and opinions on social media platforms (Veil, Buehner & Palenchar, 2011). This research will focus on the dissemination of rumours about certain events during a crisis. Therefore, in the current study, social media are defined as digital applications that facilitate interactive communication between social media users. These social media users utilize social media to seek, share and generate crisis information.

2.1.4. Social Media and Crisis Communication

Nowadays, social media have been recognised as an important crisis communication channel (Beneito-Montagut et al., 2013; Takahashi et al., 2015). According to Takahashi et al. (2015, p. 392) “social media in crisis situations have been recognized as key communication channels that can complement traditional channels”. Social media can be relevant during crisis because of several characteristics of social media. First of all, social media can be used for crisis communication due to its flat and open structure (Bruns et al., 2012). As a result, the messages shared on social media are open and visible for everyone and the flat structure allows both government authorities and citizens to seek, share and generate crisis information. Secondly, several social media platforms, such as Twitter and Facebook designed services in relation to crisis communication. For example, Twitter created ‘Twitter Alerts’: “Twitter Alerts are tweets published by select public agencies and emergency organizations during a crisis or emergency that contain up-to-date information relevant to an unfolding event, such as public safety warnings and evacuation instructions” (Twitter, 2019). Besides, Facebook developed ‘Crisis Response’. This program allows users to communicate that they are safe, give or find help, raise money and get updated crisis information (Facebook, 2019).

2.2 Functions of Social Media during Crisis

Little is known about the different functions of social media during crises (Binder, 2012). Recently, Alexander (2014) and Houston et al. (2015) identified several functions of social media during crisis. According to Alexander (2014), social media could be positively used for crisis communication in seven way. Firstly, social media can be used to listen to the public

15 debate. The listening function can reveal how people are behaving and reacting to the crisis. Secondly, social media can be used by authorities and citizens to monitor community members. Thirdly, social media can also be important for extending emergency response because social media can be used for crisis management and emergency planning. Fourthly, crowdsourcing is possible with the help of social media. Crowdsourcing enable citizens to map impacted areas and mobilise social capital and resources. Fifthly, social media can be used to enhance and create cohesion during crisis. Sixthly, social media can be used to start fundraising campaigns and collect donations. And lastly, social media can be used by researchers to understand how people use online platform during crisis (Alexander, 2014).

Houston et al. (2015) constructed a functional framework to analyse social media use during disaster. The framework consists of fifteen possible functions of social media pre-event, during event and post-event (see Annex 2). The framework is applicable to government authorities, media outlets and citizens. During the event phase, that is central for this study, only ten functions of social media are considered relevant. According to Houston et al. (2015) social media can be used positively during crisis situation to: (1) signal and detect crisis, (2) send and receive requests for help, (3) inform others about one’s own condition and location and learn about a disaster-affected individual’s condition and location, (4) learn what is happening in the disaster, (5) deliver and consume news coverage of the disaster, (6) provide and receive disaster response information, (7) donate and receive donations, (8) provide and receive support, (9) express emotions, concerns, well-wishes and (10) provide and receive information about disaster response and rebuilding (Houston, 2015).

The discussed functions of social media during crisis can positively contribute to crisis communication. However, according to Alexander (2014), mis-, dis-, and mal-information can also be disseminated on social media. Other researchers (e.g. Bruns et al., 2002; Bird, Ling & Haynes, 2012; Mendoz, Poblete, & Castillo, 2010; Starbird, Maddock, Orand, Achterman & Mason, 2014; Jong & Dückers, 2016) analysed social media use during crisis situations and they support Alexander’s (2014) claim that during crisis rumours could be disseminated on social media. Therefore, in the next paragraph, information disorder and the concepts mis-, dis- and mal- information will be discussed and defined.

2.3 Information Disorder: mis-, dis-, mal- information

During a crisis situation, not only helpful and factual information, but also rumours can be disseminated on social media (e.g. Alexander, 2014; Bruns et al., 2002; Bird et al., 2012; Jong & Dückers, 2016). A rumour can be defined as “unverified and instrumentally relevant information statements in circulation that arise in contexts of ambiguity, danger or potential threat, and that function to help people make sense and manage risk” (DiFonzo & Bordia, 2007, p.13). As a result, the dissemination of rumours could lead to information disorder and this could have serious consequences for crisis management (Wardle & Derakhshan, 2017). The concept information disorder is developed by Wardle and Derakshan in 2017, as an overarching definition that could incorporate different types of false information in a single model. They distinguish three types of information disorder, namely (1) misinformation, (2) malinformation and (3) disinformation.

According to Van Prooijen and Douglas (2017) and Van den Bos (2009), information disorder might arise in crisis situations because people want to make sense of the situation, and the aversive feelings that people experience during crises (e.g. fear, lack of control, uncertainty) may promote the development of rumours. In line with this, Wood (2018) argues that rumours and false information can reduce uncertainty and can help people make sense of a crisis situation. People in crisis situations want understandable answers and rumours (often a simplified representation of reality) can help people to understand complicated crises situations (Hofstadter, 1966; Van Prooijen & Douglas, 2017). Especially conspiracies might explain who to trust and distrust by blaming one particular group for the crisis situation (Hofstadter, 1966; Van Prooijen & Douglas, 2017). Thus, to conclude, information disorder could arise during crisis situations because uncertainty stimulates the development and believe in rumours. As a consequence, information disorder might follow a crisis situation, and social media is a platform where this could occur.

The dissemination of rumours can be harmful. Firstly, Alexander (2014) argues that the dissemination of false information (inadvertently and deliberately) could increase chaos during a crisis situation (Alexander, 2014; Castillo, Mendoza & Poblete, 2011). Secondly, the dissemination of rumours on social media can lead to an overload of information. As a consequence, it could be difficult to find credible and trustworthy crisis information (Alexander, 2014; Lin, Spence, Sellnow & Lachlan, 2016).

17 The concept information disorder is not new, but social media have changed the way crisis information is generated and distributed (Wardle & Derakshan, 2017). Formerly, crisis information was formulated by government authorities and disseminated by media outlets. However, currently, social media enable citizens to not only gather and consume crisis information, but also to generate and share crisis information (Lewandowsky et al., 2012;

Sellnow & Seeger, 2013). Besides, false information is disseminated more rapidly online (Wardle & Derakhshan, 2017). Therefore, it is important to understand how rumours are disseminated online. In this section, three types of information disorder are characterised. Wardle and Derakshan (2017) distinguish three types of information disorder, namely (1) misinformation, (2) malinformation and (3) disinformation. They distinguish information that is true from false information and information that is created and disseminated with the intent to do harm from information that is not intended to do harm. The classification of the three types are shown in Figure 1 below.

Misinformation is an information disorder that occurs when false information is shared, but without the intent to do harm. During a crisis, people could accidently share misinformation, because they want to help and do not thoroughly inspect the information before they disseminate the message (Wardle, 2018; Wardle & Derakhshan, 2017). A typical example of misinformation is a message that is misunderstood or is misinterpreted (Wardle & Derakhshan, 2017). Misinformation can be shared by government actors, media authorities and citizens. Malinformation is unverified information that is shared to cause harm. According to Wardle and Derakhshan (2017) and Marwick and Lewis (2017), people could purposefully share malinformation during a crisis for several reasons. First of all, people share malinformation because they want to promote their ideology. By disseminating malinformation this group tries to spread their ideas to the general public and tries to enhance attention for their cases. Secondly, people can gain financially from an information disorder through advertising. Thirdly, people might try to discredit authorities in an attempt to influence public opinion about responsible actors. And fourthly, some people are motivated to share malinformation to seek prestige or reinforcement (Marwick & Lewis, 2017; Wardle & Derakhshan, 2017). A typical example of malinformation is a conspiracy theory, in which an event is explained as the result of a group of people or higher power who are secretly cooperating with evil intentions (Birchall, 2006).

Disinformation is false information that is deliberately created and disseminated to cause harm. Disinformation includes elements of both misinformation and malinformation. Specifically, disinformation includes messages that use a false context, imposter content, and manipulated or fabricated content (Sellnow et al., 2019; Wardle & Derakshan, 2017). Producers of disinformation typically have political, financial, psychological, or social motivations to share false information and create harm (Wardle, 2018). An example of disinformation is a hoax, in which a false claim about an organisation, person or group is disseminated to influence their reputation (Sellnow et al., 2019).

Figure 1: The Classification of Information Disorder (Wardle and Derakshan, 2017).

The taxonomy is developed by Wardle and Derakhshan in 2017. The taxonomy is still rather new and not completely worked out yet. Besides, the three models do not entirely exclude each other. However, the taxonomy is mainly intended as a sensible concept and can be helpful to put some order in the amount and variety of concepts regarding information disorder. In the next paragraph, the self-correcting mechanisms thesis and the official dominance thesis that might explain the correction of false crisis information on social media will be further examined.

19 2.4 The Self-Correcting Mechanisms Thesis

According to Jong and Dückers (2016), the self-correcting mechanisms thesis can help explain the correction of rumours on social media during crisis. The self-correcting mechanisms thesis argues that social media communities are able to correct rumours with the help of three mechanisms. It would be interesting to analyse if the self-correcting mechanisms could also explain the potential correction of rumours during the Utrecht tram shooting.

On the one hand, according to Wardle and Derakhshan (2017), on social media, false information is disseminated between trusted peers. Therefore, mis-, dis-, and mal- information might not easily be questioned or corrected on social media because like-minded people in a group can become an ‘echo chamber’ (Sunstein, 2002; Wardle & Derakhshan, 2017). In an echo chamber, users can constantly choose to read information in line with their beliefs and disseminate their perspectives among like-minded people without any criticism (Wardle and Derakshan, 2017). As a result, echo chambers on social media may lead to ‘selective exposure’ and ideological segregation (Flaxman, Goel, & Rao, 2016; Sunstein, 2002; Wardle & Derakshan, 2017, p.50). During crisis, social media users can also disseminate rumours (e.g. hate, harassment and hoaxes) in echo chambers. As a result, false (crisis) information may not be criticised and the false information in echo chambers may become more extreme (Sunstein, 2002).

On the other hand, several researchers have also indicated that social media communities have the ability to correct mis-, dis- and mal- information during crisis (Bruns et al., 2002; Bird et al., 2012; Mendoza et al., 2010; Starbird et al., 2014; and Jong & Dückers, 2016). For example, Bruns et al. (2002), Bird et al. (2012) analysed the use of Facebook during the Queensland and Victorian floods. They concluded that during the floodings rumours were common, but they argued that the Twitter account @QPSMedia’s was active during the crisis to correct rumours and the tweets of @QPSMedia’s were extensively retweeted. As a result, the use of social media during the crisis did not lead to the spread of mis-, dis-, and mal- information. In line with this, Mendoza et al. (2010) analysed the earthquake in Chili in 2010. They concluded that on social media rumours were questioned much more often than confirmed truths. This could indicate that social media communities try to validate information and try to correct false information. Besides, Starbird et al. (2014) analysed the Boston marathon bombing and discovered that rumours were corrected on social media. However, they also state that the proportion of corrections was very small in comparison to the dissemination of the rumours during crisis.

Recently, Jong and Dückers (2016), more specifically analysed the correction of rumours on Twitter during an incident in the Netherlands, in which a gunman entered a news station and demanded airtime. They concluded that during the crisis two rumours were spread intensively on Twitter namely: (1) an identical situation took place in Belgium and (2) the parents of the gunman were killed in the MH17 tragedy (Jong & Dückers, 2016). Jong and Dückers (2016) determined that social media platforms did not only stimulate the dissemination of these rumours, but they also concluded that users were able to correct rumours with the help of self-correcting mechanisms.

Jong and Dückers (2016) identify three self-correcting mechanisms that social media communities can use to correct rumours during crisis. First, the sender of the incorrect tweet can try to correct the rumour. A sender can correct rumours during crisis by sending a follow-up tweet with corrected information or by deleting the tweet with the false information. Second, other Twitter users can try to validate the information and can correct false information. For example, the Twitter community can confirm, dispute or question rumours during crisis. Third, Jong and Dückers (2016, p.340) identified a ‘wisdom of crowds’ in which users on social media can help further disseminate or challenge the corrected messages on social media during crisis. For example, the community can help disseminate the corrected information and can retweet the correction to increase the reach of the corrected information. The community can also challenge the corrected messages by questioning and validating the corrected information. 2.5. The Official Dominance Thesis

According to Korthagen (2015) and Shehata (2010), the official dominance thesis can help explain the dissemination of information through traditional media. It would be interesting to analyse if the official dominance thesis could also explain the dissemination of crisis information on social media.

The official dominance thesis argues that official actors, such as government authorities and politicians often get more media attention. According to Korthagen (2015, p.61), “the more power an actor holds, the more media attention he/she automatically receives in the media.” As a result, authorities will dominate the news and get more media coverage (Korthagen, 2015; Shehata, 2010). Officials dominate the news because of journalistic expectation of objectivity. Journalists often look at officials because they expect that officials disseminate ‘factual, authoritative and legitimate information’ (Korthagen, 2015, p.61).

21 This could also indicate that government authorities and their messages get more coverage on social media during crisis because their messages are more disseminated than other sources because they are trusted sources of factual, authoritative and legitimate crisis information. Besides, if authorities correct rumours, it could be expected that this message will get media coverage on (social) media. For example, authorities may have more followers and messages from authorities may get more retweets.

Lin et al. (2016) state that it is very important that government messages get enough media coverage on social media during crisis. Besides, Lin et al. (2016) argue that government authorities have an important role in correcting and monitoring rumours on social media. Therefore, government authorities should provide citizens with enough and credible crisis information. To accomplish this, government authorities can promote their own hashtag. Authorities may also engage in a two-way dialogue with the public on social media to disseminate credible information and correct rumours.

2.6. Research Expectations

To conclude, social media are an important crisis communication channel during crisis (Beneito-Montagut et al., 2013; Takahashi et al., 2015). On social media, users can seek, share and generate crisis information (Veil et al., 2011). This poses challenges, because on social media not only helpful and factual information is shared, but also mis-, dis-, and mal- information can be disseminated and this could further disrupt a crisis situation (Wardle & Derakhshan, 2017).

First, the self-correcting mechanisms thesis argues that social media communities might be able to correct rumours on social media during a crisis situation. Besides, the thesis identifies three mechanisms that could be used during crisis to correct rumours: (1) the sender of the incorrect tweet can try to correct the information, (2) other social media users can try to validate the crisis information, and (3) social media users can help disseminate or challenge the corrected messages. Secondly, the official dominance thesis, suggests that authorities could get more attention on social media during crisis. This could indicate that crisis (correction) messages of authorities could get more coverage on social media during crisis (retweets and followers) because they are trusted sources of factual, authoritative and legitimate crisis information.

Based on aforementioned theories two research expectations (H) can be formulated:

The Self-Correcting Mechanisms Thesis:

H1: Social media communities are able to correct mis-, dis-, and mal- information on social media during the Utrecht tram shooting because (1) the sender of the incorrect tweet tries to correct the information, and/or (2) other social media users try to validate the crisis information, and/or (3) social media users help to disseminate or challenge the corrected information. The Official Dominance Thesis:

H2: Messages that aim to correct mis-, dis- and mal-information dominate social media during the Utrecht tram shooting, if the source of this message is an official government authority on Twitter.

23

3. Methodology

The aim of this study is to answer the question: “What mechanisms explain the dissemination and correction of rumours on Twitter during the tram shooting crisis in Utrecht?” To answer this research question, choices were made regarding the methodology and choices have to be justified. This chapter will therefore elaborate on the research strategy, research design, case selection, data collection, data analysis, research ethics and research limitations.

3.1 Research Strategy

A mixed-method research design was used. With the help of qualitative research, the rumours on social media during the crisis were examined. Subsequently, quantitative data from Twitter (e.g. number of followers, retweets and responses) was analysed to understand the reach of false crisis information on social media (Bryman, 2015). The objective of this research was to understand the dissemination and correction of rumours and the role of authorities therein. Therefore, I analysed (1) six rumours that were present on social media during the Utrecht tram shooting. I studied (2) how mis-, dis-, and mal-information was corrected on Twitter (correcting-mechanisms thesis). And (3) I explored which actors were involved in the discussion on Twitter (official dominance thesis) was analysed.

3.2 Research Design: A Single Case Study

To answer the research question a single case study was conducted regarding the Utrecht tram shooting on March 18th, 2019. I purposefully chose for a case study research design. First, a case study could provide an in-depth and nuanced understanding of this real-life social event, namely the dissemination and correction of rumour on social media (Bryman, 2015; Van Thiel, 2015; Yin, 2018). Secondly, a case study design is applicable to exploratory research and may yield valuable exploratory results (Resodihardjo, 2016; Yin, 2018). This research was exploratory because little research was available about the possibility of social media communities to correct rumours on social media during crisis. Thirdly, a case study research design enabled me to test and further develop the existing theoretical concepts, namely the self- correcting mechanisms of social media communities and the official dominance thesis (Eisenhardt, 1989; Yin, 2009).

3.3 Case Selection

To answer the research question, I analysed tweets related to the tram shooting in Utrecht on March 18th_{, 2019. The Utrecht tram shooting was chosen because it is a typical case for the} broader phenomenon I would like to study, namely the dissemination and correction of rumours during crisis situations (Bryman, 2015; Yin, 2009). A typical case study can "describe and illustrate what is typical" (Patton, 1990, p.173). This is interesting because little research is available about the Utrecht trams shooting case. The case is typical because I wanted to answer the research question: “What mechanisms explain the dissemination and correction of rumours on Twitter during the tram shooting crisis in Utrecht?” First, the tram shooting in Utrecht is a typical case because the event can be classified as a crisis as it evokes feelings of threat, uncertainty and urgency by both government authorities and citizens (Boin, 2005). Second, the case is a typical case because the crisis stimulated social media use by government authorities and citizens. Third, social media gave rise to the dissemination of mis-, dis-, and mal- information on Twitter. This last point; the dissemination of mis-, dis-, and mal- information, in particular made this case a typical case for my research. Because the aim of this study was to understand the dissemination and the correction of rumours on social media during crises. This case was chosen because the crisis occurred recently and has been extensively discussed on social media platforms. This made data collection possible and relevant. Besides, the majority of tweets are written in Dutch. As a result, I am able to read and understand the tweets. Tweets related to the event were analysed since the start of the shooting at 10:45 until 18:15 when the perpetrator was taken into custody. The scope of the analysis is limited to this timeframe to understand the impact of false information on a crisis situation.

3.4 Within-Case Analysis: mis-, dis-, and mal- information

To answer my research question, I conducted a within-case analysis. A within-case analysis was chosen because I wanted to get an in-depth understanding of the use of social media during the Utrecht tram shooting. Within the single-case study (the Utrecht tram shooting), several sub-units were distinguished related to: (1) misinformation (2) disinformation and (3) malinformation. Six (false) rumours were extensively analysed related to the sub-units of false information (see Table 1). The analysis of sub-units enhanced the number of analyses within the single-case study. This improved internal validity because an in-depth understanding of the case study was possible. A within-case analysis made generalisation within and across the

sub-25 units possible (Vaughan,1996; Yin, 2003). See Table 1 for the rumours that were analysed in this research report.

Type of Information Disorder Rumour Utrecht Tram Shooting

Misinformation:

False information disseminated without a harmful intent (Wardle & Derakshan, 2017)

1. There are multiple shooting sites 2. There is a shooting at a Mosque

Disinformation:

False information disseminated with a harmful intent (Wardle & Derakshan, 2017)

1. Suspect in Utrecht published manifesto online 2. Militant Dutch white nationalist, Sam Hyde,

claims responsibility Malinformation:

Genuine or unverified information disseminated with a harmful intent (Wardle & Derakshan, 2017)

1. Turkish and Moroccan boys celebrate the tram shooting

2. The tweet by fake account @PolitieP: All mosques closed

Table 1: Selected Rumours Utrecht Tram Shooting.

The rumours were selected inductively. First, the tweets that were sent during the crisis were widely read in order to get a comprehensive overview of the rumours that were present during the crisis. Second, as an extra check, Coosto was used to search for rumours with keywords such as ‘hoax’, ‘false information’ and ‘rumour’. Eventually, the six rumours were selected because they were retweeted or discussed a lot on Twitter during the Utrecht tram shooting. Besides, rumours were selected because they could be characterised as either mis-, dis-, and mal- information. As a result, the different rumours can be analysed and compared within and across the different subunits.

3.5 Social Media Platform Twitter

To answer the research question, data was gathered from Twitter. Twitter is chosen as social media platform, to understand the use of social media by government authorities and citizens during the crisis. The research is limited to the social media platform Twitter because Twitter is shown to be the preferred platform for crisis communication due to the flat and open structure of the network (Bruns et al., 2012). As a consequence, tweets on Twitter are open and visible for everyone and the flat structure allows authorities and citizens to both gather, share and produce information.

To analyse the dissemination of rumours on social media platforms it is first of all important to explain the social media platform Twitter. Twitter is a free social media platform. People with an account on Twitter are first of all, able to write tweets and share them on their own platform. These messages can contain a maximum of 250 characters. Secondly, users can follow a variety of accounts on Twitter. For example, users can follow friends, family, celebrities and authorities. When Twitter usage follows other users, they are able to read their tweets. Thirdly, users can retweet other users’ tweets and share them on their own platform. As a result, tweets can reach a bigger audience. Lastly, Twitter allows users to respond to other tweets. Consequently, on the one hand, users can express support for specific tweets, ideas and opinions. On the other hand, Twitter users are also able to criticise or question tweets from other users.

To summarise, during a crisis situation, Twitter users can write messages about the event. These messages are disseminated to their followers. Besides, users can retweet information about the event and thereby increase the reach of these tweet. And lastly, Twitter users can respond to other tweets during crisis to confirm, criticise or question rumours on social media during crisis. 3.6 Self-Correcting Mechanisms: Operationalisation

In this research report I wanted to empirically assess the self-correcting mechanisms thesis with a social media analysis consisting of six rumours. The self-correcting thesis argues that rumours can be corrected on social media during crisis. The operationalisation of Jong and Dückers (2016) was used to analyse the concept and to understand whether rumours were corrected on social media. Jong and Dückers (2016) coded tweets as either irrelevant (when tweet did not discuss the rumour), question (when tweet questioned the rumour), true (when people confirmed the rumour) or false (when people denied the rumour). I have elaborated the operationalisation of Jong and Dückers (2016) and applied this to the Utrecht tram shooting. Besides, Jong and Dückers (2016) have identified three mechanisms that social media communities may use to correct rumours. They have not operationalised these three mechanisms, but with the help of their theory I have operationalised the mechanisms. See Table 2 for the operationalisations and see Annex 1 for the codebook.

27 Table 2: The Operationalisation of the Self-Correcting Mechanisms Thesis for the Utrecht Tram Shooting.

Code Categories Variable Indicator

1 Confirm

Rumour

The rumour is confirmed on Twitter.

• The tweet is retweeted (It is not possible to trace if a retweet really means that the person agrees with the tweet. But in this case, it is coded as such because the person wants the tweet to be disseminated).

• A reaction to the tweet in agreement with the rumour.

2 Dispute

Rumour

The rumour is disputed on Twitter.

• A reaction to the tweet that says that the rumour is incorrect/false.

• A reaction to the tweet entails evidence against the rumour.

3 Question

Rumour

The rumour is questioned.

• A reaction to the tweet entails a question to other Twitter users about the correctness of the rumour.

• A reaction that asks for evidence to support the claim. • A reaction to the tweet suggests other options to explain

the event.

4 Irrelevant to

Rumour

The rumour is not mentioned in the tweet.

• The tweet is about another debate.

• The tweet is too short to identify other categories. • The tweet is written with only emoticons or written in a

different alphabet. A Self- Correcting Mechanism 1: The Sender of the Rumour

The sender of the incorrect tweet can try to correct the rumour.

• The sender sends a follow-up tweet with the corrected information.

• The sender deletes the tweet with the false information.

B Self- Correcting Mechanism 2: The Twitter Community

Other Twitter users can try to validate the information and can correct false information.

• Twitter users can dispute the tweet with the rumour. • Twitter users can question the tweet with the rumour.

C Self- Correcting Mechanism 3: Wisdom of Crowds

Other Twitter users can help to

disseminate or challenge the corrected messages.

• The community can retweet the corrected information to increase the reach of the corrected information.

• The community can challenge the corrected information by validating/questioning the corrected information.

3.7 Official Dominance: Operationalisation

In this research report I wanted to empirically assess the official dominance thesis with a social media analysis consisting of six rumours. According to Korthagen (2015, p.61), “the more power an actor holds, the more media attention he/she automatically receives in the media”. This indicates that authorities and their messages may dominate the discussion on social media during crisis situations (Korthagen, 2015; Shehata, 2010). During the Utrecht tram shooting, government authorities, media authorities, politicians, journalists, influencers and citizens disseminated crisis information on social media. According to the official dominance thesis, tweets from government authorities may dominate social media because these accounts have a lot of followers and are retweeted a lot (trusted sources of legitimate information). The operationalisation of Korthagen (2015) was used to analyse the concept and to understand whether (correction) messages of authorities dominated the discussion of the rumours on social media. I have elaborated on the operationalisation of Korthagen (2015) and applied this to the Utrecht tram shooting. See Table 3 for operationalisation and see Annex 1 for the elaborated codebook.

Code Categories Variable Indicator

a Government

Authority

Any governmental entity (administrative or regulatory body) that disseminates information on Twitter.

• Tweets sent by the police • Tweets sent by the NCTV • Tweets sent by province • Tweets sent by municipality • Tweets sent by university

b Media

Authority

Any media entity engaged in

disseminating information on Twitter to the general public (for example through a newspaper, radio, television, or other medium of mass communication).

• Tweets sent by newspapers (e.g. Volkskrant)

• Tweets sent by news broadcasting agencies (e.g. NOS, TV24) • Tweets sent by online

newspapers (e.g. The Post Online)

c Personal

Account Politician

A person who is professionally involved in politics and writes tweets on their personal account on Twitter.

• Tweets sent by people who state that they are politicians in their Twitter profile.

29

d Personal

Account Journalist

A person who writes for newspapers, magazines, or news websites or prepares news to be broadcast that write tweets on his personal account on Twitter.

• Tweets sent by people who state that they are journalists or freelancers in their Twitter profile.

e Influencer A citizen with an exceptional number of

followers on Twitter. As a result, the person’s influence on Twitter can be considered large.

• Tweets sent by citizens with more than 3.000 followers.

f Personal

Account Citizen

Anyone who does not fall into the aforementioned categories.

• Tweets sent by citizens.

Table 3: The Operationalisation of the Official Dominance Thesis for the Utrecht Tram Shooting.

3.8 Data Collection

The search terms #24oktoberplein and #Utrecht were used to collect relevant tweets. These search terms were chosen because in the official government communication these hashtags were used to refer to the crisis situation and other people copied these hashtags to discuss the crisis.

During the Utrecht tram shooting (18th _{March: 10:44-18:15) 44.899 tweets were sent with the} hashtags Utrecht or 24oktoberplein. To make the data analysis feasible only the tweets that related to the six (false) rumours were analysed. As a result, the analysis was limited to 1.081 tweets. In Table 4 the number of tweets per rumour are distinguished.

Rumour Utrecht Tram Shooting Number of analysed Tweets (n= 1.081)

Misinformation 1: There are multiple shooting sites 206 Misinformation 2: There is a shooting at a Mosque 72 Disinformation 1: Suspect in Utrecht published

manifesto online

224

Disinformation 2: Militant Dutch white nationalist, Sam Hyde, claims responsibility

43

Malinformation 1: Turkish and Moroccan boys celebrate the tram shooting

Malinformation 2: The Fake Account of @PolitieP: All Mosques Closed

197

Table 4: Collected Tweets per Rumour.

The social media analysis of the different rumours consisted of different parts. First, to understand the general use of social media during the crisis I:

- Identified which hashtags were used to communicate during the crisis. - Identified how many tweets were sent with these hashtags during the crisis.

- Identified how many users participated on Twitter with these hashtags during the crisis. - Identified which accounts were most visible during the crisis.

Second, to understand the correction of misinformation, disinformation and malinformation on social media I:

- Identified how rumours were disseminated on Twitter during the tram shooting crisis in Utrecht? (related to sub-research question 1)

- Identified whether there were efforts within the Twitter community to correct the rumours. Therefore, tweets were coded: confirm, dispute, question or irrelevant (related to the correcting-mechanisms thesis, sub-research question 2, and expectation 1).

- Identified which mechanisms were used on Twitter to correct the rumour. Tweets that corrected the rumour were coded: the sender of the rumour corrected the false information, the Twitter community corrected the rumour, or a wisdom of crowds further disseminated or questioned the corrected information (related to the correcting-mechanisms thesis, sub-research question 2, and expectation 1).

- Identified who tried to correct mis-, dis, and mal-information. These tweets were coded: government authorities, media authorities, personal account politician, personal account journalist or personal account citizen (related to the official dominance thesis, sub-research question 3, and expectation 2).

3.9 Data Analysis

The Software program Coosto was used to search and collect the tweets related to the six rumours during the Utrecht tram shooting. Coosto is a tool that enables organisations to analyse their reputation on social media platforms. Besides, Coosto enables academics to collect data for a (qualitative and quantitative) social media analysis (Jong & Dückers, 2016). With the help

31 of Coosto, I was able to collect all tweets within a specific timeframe and with specific keywords. Tweets were collected that contained the hashtags #24oktoberplein and #Utrecht to get a general understanding of social media use during the crisis. Thereafter, Coosto was used to collect all the tweets that related to a specific rumour. The discussion related to the rumour was completely uploaded to Excel.

The use of Coosto entailed a number of advantages and disadvantages. On the one hand, Coosto enabled me to download all the tweets related to one particular discussion. This ensured that all tweets that responded to the rumour were included in this analysis. On the other hand, with Coosto only tweets that responded to Twitter users that were part of the discussion could be collected. Tweets that discussed the rumour without immediately responding to the discussion on Twitter could not be collected. Besides, with Coosto it is only possible to download tweets from accounts that still exist on Twitter. Deleted accounts and their tweets cannot be retrieved. As a result, the analysis of the rumours relating to disinformation are incomplete. The complete timeline and the discussion cannot be analysed because the original accounts that started the rumour have been deleted.

To understand the dissemination and the correction of rumours on social media during crisis, a qualitative content analysis was conducted to assess the gathered data. Coding helped interpret and theorise the collected data (Bryman, 2015). Besides, coding made comparison between and within sub-unities of analysis possible (Bryman, 2015). Excel was used to code the data (Bryman, 2015). Codes were derived from existing theory (e.g. Jong & Dückers, 2016). Eventually, a codebook was developed that is adapted to the Utrecht tram shooting. The codebook and a coding example can be found in Annex 1.

3.10 Research Ethics

Usernames, occupations, and tweet messages of Twitter users were included in this study. The privacy of Twitter users was not violated, because the Twitter accounts were publicly available.

3.11 Research Limitations

In this section the reliability and validity of this research report are examined based on the criteria of Guba and Lincoln (described in Bryman, 2015). These criteria are used because they are specifically developed for qualitative research and thus also applicable to mixed-method

research designs. The criteria are credibility, transferability, dependability and confirmability. Credibility refers to internal validity, transferability to external validity, dependability to reliability, and confirmability to objectivity (Bryman, 2015).

Credibility

Credibility is about assessing the extent to which the results of a study are credible. The main issue is whether the research gives a good understanding of reality (Bryman, 2015). According to Bryman (2015), the credibility of research can be guaranteed with the help of respondent validation and triangulation. To answer the research question a social media analysis was conducted. Therefore, respondent validation and triangulation were not guaranteed. However, the internal validity of the research is high, because the case was analysed in-depth. Therefore, there is a good understanding of the context of the crisis what enhances credibility (Yin, 2018). Transferability

Transferability is about assessing the extent to which the results of a study also apply to other contexts (Bryman, 2015). This research report focussed on the tram shooting in Utrecht and on the social media platform Twitter. On the one hand, this ensured that the case was analysed in-depth. On the other hand, the generated results from the unique case might be difficult to generalise (Bryman, 2015; Van Thiel, 2015; Yin, 2018). First, in order to make the results relevant in a broader context, thick descriptions have been used (Bryman, 2015). With the help of thick detailed descriptions of the results, interested parties can judge for themselves whether the results apply to other crises as well (Bryman, 2015). Second, to make generalisation possible, I focused on “analytic generalisation” (Van Thiel, 2015, p.104; Yin, 2003, p.31-32). Previous scholars have generated theories about the possibility of social media communities to correct rumours. I have studied whether these mechanisms could also explain the dissemination and the correction of rumour on social media during the Utrecht tram shooting. Therefore, this study may have tested and further developed existing theories and enhanced analytic generalisation (Eisenhardt, 1989). Third, I have created sub-units of analyses to multiply the amount of observations. This enhanced generalisation between and across the different sub-units (Yin, 2003). Lastly, I acknowledge that this study will not generate definite answers applicable to the dissemination of rumours on social media during all crises. However, this research might generalise theoretical propositions and could start a debate about the dissemination of rumours on social media and the possibility of social media communities to

33 Dependability

Dependability is about assessing the extent to which the results of the research are influenced by the bias of the researcher (Bryman, 2015). For a study it is important that the dependability is low to enable other researchers to achieve the same results with the same methods. To enhance reliability, I have tried to conduct this research in a systematic way. Therefore, the core concepts are clearly operationalised and defined, the collection of tweets are explained, and a coding scheme is developed. As a result, other researchers can follow the same method, find the same tweets and find the same results.

Confirmability

Confirmability is about the objectivity and neutrality of the research (Bryman, 2015). This research report focussed on the interpretation of tweets. As a consequence, the findings may be affected by personal interpretation. This risk is minimalised with an attentive open attitude to prevent an observation bias (Bryman, 2015). Besides, I have asked two students to code some tweets related to a rumour. This helped me to see if there was consistency between interpretations of tweets and helped me to optimise the code sheet.