3. Method
3.1. Data Collection
3.1.1. Twitter Content Analysis
For the data mining, this study utilized the Python library Tweepy to connect directly to Twitter streaming API. This study focused on tweets using the hashtag #KamiTidakTakut that emerged quickly after the attack and was consistently used for the KM bombing incident. The duration of tweet collection is from May 24, 2017, when KM bomb detonated until June 5, 2017, when the conversations subsided. The total number of tweets collected is 7,685. Python library Tweepy is also able to retrieve the metadata of the tweets which will be useful for this study, for instance the username, date, numbers of retweets, numbers of favorites, texts of the tweets, locations, mentions, hashtags, and permalinks.
DOI:10.6814/THE.NCCU.IMICS.006.2018.F05 During the preliminary analysis, this study identified that most of the tweets were in Indonesian and a small amount were in English, French, Japanese, Thai, and Korean. As the coders were able to understand multiple languages, all tweets in Indonesian and English were analyzed directly while tweets in other languages were analyzed after translation. After further identification, some tweets in Japanese and in Thai were not posted by Indonesians and as this current study attempted to explore how Indonesians responded toward the terrorist attack, the tweets posted by non-Indonesians were eliminated from the analysis. The preliminary analysis also identified irrelevant tweets from the dataset. Irrelevant tweets to the KM bombing were mostly advertisements, discussions about other topics, or jokes. In total, there were 584 tweets removed after identifying the language of the tweets and irrelevant tweets, leaving 7,101 tweets for further analysis.
This study also identified the Twitter mobilizers based on their engagement rate (retweet, reply, favorite). LeFebvre and Armstrong (2016) distinguished four types of digital participants on social media participation during protests or emergency: key mobilizers, unwitting mobilizers, moderate influencers, and passionate participants. In Twitter content analysis, this study focused more on identifying the most influential users based on their engagement rate and listed the top 15 mobilizers which mostly belong to the categories of key mobilizers and unwitting mobilizers. This study also utilized the examination of the digital participants to recruit the interviewees for the next step of the study. However, most participants belonging to these categories were mostly the government institutions/ individuals or security forces while this study would like to understand the dual screening experience of the public’s dual screening use. Thus, this investigation also identified other digital participants who used dual screening to discuss about KM bombing attack to be recruited for the in-depth interviews.
DOI:10.6814/THE.NCCU.IMICS.006.2018.F05 3.1.2. In-depth Interview
This research conducted in-depth interviews with the different affected groups: incident-related actors (survivors and their family members/friends and affected residents), digital participants, and journalists who covered the terrorist attack. In total, there are 21 Indonesian dual screeners from the three aforementioned groups interviewed in this study. Each group consisted of seven respondents. There are some other criteria to be fulfilled by the participants. All respondents must be active dual screeners who own at least two screen devices (TV, smartphone, laptop, tablet, and so on). Moreover, they should have prior experiences of dual screening use during the KM bombing case in May 2017. Users aged 20 to 35 are most active dual screeners (SWAonline, 2017) and major social media users in Indonesia (Emarketer, 2016), which is the age selection criterion for interviewees.
The purposive and snowball sampling was employed to recruit the respondents of this study. However, there are different recruitment methods for each respondent group for recruiting the digital participants, the first-step Twitter analysis was used to identify dual screeners on Twitter. As previously mentioned, the in-depth interview recruited digital participants who discussed KM bombing attacks. The most influential users on Twitter such as the top 15 key mobilizers were primarily government-related institutions or individuals, thus they did not fulfill the criteria. Interview invitations were sent to Twitter users belonging to these categories through Twitter direct messages or tweets. Moreover, for the journalists group, the researcher observed the online articles or videos about the KM bombing that appeared on Twitter and contacted the reporters to be interviewed. These journalists then referred other journalists who also had reported the incident. To assure the variability, each journalist could only refer another person. These
DOI:10.6814/THE.NCCU.IMICS.006.2018.F05 journalists work for different media, including TV, online news portals, and newspapers. Lastly, to recruit the incident-related actors, the researcher observed activities on social media and also announced the recruitment information on these platforms. People who responded to the advertisement were contacted to schedule the interview.
Prior to the interview, all respondents must sign the consent form (see Appendix A). They were allowed to terminate the interview process if they felt uncomfortable. All collected data from the interviews was treated as confidential information as this study only used personal data for research purposes only. To protect the personal data, the respondents of this study were under name codes (I for incident-related actors, D for digital participants, J for journalists). There were 34 questions for the interview to answer the three proposed research questions and covered four aspects: filtering questions, demographic questions, motivations of dual screening use, and crisis responses of terrorist attacks. The questions were also translated into Indonesian by the researcher and another Indonesian speaker was recruited to recheck the translation to ensure the meaning was not altered (see Appendix B). Each interview took place for around 30 to 60 minutes and the conversations were audio recorded for transcription and analysis. This research had the interview guides in English and Indonesian. All interviews transcripts in Indonesian were translated into English. Along with the researcher, another Indonesian speaker was recruited to recheck the translation in order to ensure the meaning of the conversations was not lost or altered in translation.
The participants were offered IDR 50,000 phone credit as the incentive.
DOI:10.6814/THE.NCCU.IMICS.006.2018.F05 3.2. Data analysis
3.2.1. Twitter data analysis
There are at least seven steps in the process of content analysis, including “formulating the research questions to be answered, selecting the sample to be analyzed, defining the categories to be applied, outlining the coding process and the coder training, implementing the coding process, determining trustworthiness, and analyzing the results of the coding process” (Shieh & Shannon, 2005, pp. 1285). To develop the code scheme, this study employed a hybrid approach, combining inductive and deductive approach to the coding (Fereday & Muir-Cochrane, 2006). By implementing this approach, the code scheme has a balance of existing framework (deductive approach) and emerging codes from the data (inductive approach). The deductive coding was adapted from the theoretical framework. The inductive approach was used to identify emerging codes during data analysis.
This research adapted the codes from Heverin and Zach (2010) who argued that individuals use Twitter to manage and disseminate information related to the crisis and categorized crisis-related Twitter contents into emotion-crisis-related, information-crisis-related, and action-crisis-related messages.
Emotion-related content is further developed three sub-codes (emotional venting, offering prayer, and expressing sympathy), based on Bautista and Lin (2015). The fourth sub-code, expressing solidarity, was identified from the data analysis. Bautista and Lin (2015) also categorized informational tweets into some sub-codes, such as posting or sharing crisis pictures or videos, news, personal comments, and status updates. The sub-codes of action-related tweets (i.e., calling to unite, calling to stop posting sensational pictures or videos, reminding to stay safe, and promoting enforcement of anti-terrorism law) were emerged after the analysis. Table 1 shows the
DOI:10.6814/THE.NCCU.IMICS.006.2018.F05 final code scheme of the Twitter content analysis. However, although each tweet could only have 140 characters, one single tweet might contain a lot of messages. There were several tweets in this study that contained more than one sub-code in one tweet and therefore these tweets were coded to multiple sub-codes as long as they could fulfill the criteria of the operational definition.
Prior to coding the whole dataset, 300 tweets were randomly retrieved from the pool to test the intercoder reliability. Along with the researcher, an Indonesian speaker was recruited and trained to independently code the tweets based on the code scheme. The Cohen Kappa from this preliminary test was 0.899, implying acceptable intercoder reliability. The disagreements from this preliminary test were reviewed for better agreement rate on the rest of the data.
Table 1
Code Scheme for Twitter Data Analysis
Code Operational Definition Sample Tweet
Action-related
Tweets for calling an action, such as telling other individuals to be safe, to praise law enforcement, and to annotate online profiles (Heverin & Zach, 2010)
Calling to unite Asking other Indonesians to
unite to counter terrorism Mari bersatupadu melawan teroris ISIS #KamiTidakTakut [Let’s and the videos of the victims or the event
Stop share the victim's photos! It's not funny #PrayForJakarta
#KamiTidakTakut Reminder to stay safe Asking other Indonesians be
more cautious and stay safe Everyone please stay safe
#PrayForJakarta
DOI:10.6814/THE.NCCU.IMICS.006.2018.F05 Emotion-related
Tweets containing emotional content such as personal emotional statements (Heverin &
Zach, 2010) Offering prayer Offering prayers for the
victims and the country #PrayForJakarta May the soul of the victims rest in peace
#KamiTidakTakut
#KampungMelayu
#WeStandTogether Expressing sympathy Expressing condolences to
victims Expressing solidarity Expressing the solidarity or
support for the police and the country
Tweets contain information sharing activities (Heverin & Zach, 2010) Posting or sharing news Sharing news stories related
to the Kampung Melayu
The terrorism that Isis had done is NOT what Islam teach us, all religions teach us to loves others
#KamiTidakTakut
#manchesterattack
DOI:10.6814/THE.NCCU.IMICS.006.2018.F05 Providing status updates Real-time posting of
activities and observations
2 people confirmed died in the suicide bomb blast, at bus interviews. This analysis technique is able to find, analyze, and describe the patterns within the data (Braun & Clarke, 2006) and is suitable for interpretation (Vaismoradi et al., 2013). Braun and Clarke (2006, pp. 16-23) also suggest six phases in analyzing the data, “1) familiarizing yourself with the data, 2) generating initial codes, 3) searching for themes, 4) reviewing themes, 5) defining and naming themes, and 6) producing the report.” As what has been described on the phases, thematic analysis often focuses on codes and themes (Marks & Yardley, 2004).
This study identified four codes and 14 sub-codes based on previous literature review and research questions and one code that emerged after the analyses (see Table 2). In response to RQ2, this study adapted the CRCM framework with four phases of crisis response process (observation, interpretation, choice, and dissemination) (Hale et al., 2005) and made an adjustment to add the code “connectivity” as an underlying activity to fit the context of dual screening use in social media
DOI:10.6814/THE.NCCU.IMICS.006.2018.F05 and video viewing (Dorn et al., 2007). In order to answer RQ3, many existing literatures have examined the social, cognitive, and affective factors to use dual screening. According to the dual screening study by Lin and Chiang (2017), social factors (social presence and bridging social capital) were identified as predictors influencing dual screening use which influenced online and offline political participation. Shin (2013) also found sociability as a predictor affecting intention to use STV. Social change was also found as a social factor to use dual screening after analyzing the interview data. Gil de Zuniga and colleagues (2015) suggested two cognitive motives:
information seeking and information appraisal. Additionally, Chadwick et al. (2017) also identified information sharing as one of the cognitive factors. Hence, this present study adapted these three sub-codes for cognitive motives (e.g., information seeking, information appraisal, and information sharing). Lastly, the sub-codes for affective motives (e.g., emotional support and emotional venting) were adapted from Jin (2010).
Table 2.
Code Scheme for Interview Data
Codes Subcodes Definition Reference
Crisis Response Process
Observation Gathering data regarding unfolding
crisis events Hale, Dulek,
Hale, 2005 Interpretation Assigning meaning to those ongoing
crisis events observed in the
observation and assessing information within the context of the current crisis to determine both its accuracy and its relevance
Hale, Dulek, Hale, 2005
Choice Examining the overall picture of the crisis situation that emerged from the interpretation step, discuss the viability of candidate action
Hale, Dulek, Hale, 2005
DOI:10.6814/THE.NCCU.IMICS.006.2018.F05 alternatives, and choose those
alternatives to implement
Dissemination Information exchange with the public Hale, Dulek, Hale, 2005 Connectivity A need to communicate with his or
her friends over social media and hence suggests it to others.
Quan-Haase &
Young, 2010
Motivations
Social motives Social presence Psychological variable that goes beyond the virtual presence of other social actors and generates subjective
capital Providing useful information or new perspectives for one another between within a social network, derived from conversations and communications on social media
Lin & Chiang, 2017
Sociability How the technical features (technological affordances and
Social Change Insisting social change in response to
the collective disadvantage Van Zomeren et al., 2004
Cognitive motives Information seeking
Getting up-to-date information. Gil de Zuniga et al., 2015
Information
appraisal Obtaining additional information about the content they are watching beyond what the video offers and to clarify unknown information
Han & Lee, 2014
DOI:10.6814/THE.NCCU.IMICS.006.2018.F05 Information
sharing
Spreading information to online communities or networks
Stieglitz &
Dang-Xuan, 2013
Affective motives Emotional support
Exchange between individuals that involve expressing care and affection, validating the person’s worth, and confiding (listening to problems)
Lincoln, Taylor
& Chatters, 2013
Emotional venting
Expressing anger and other negative emotions
Parlamis, 2012
In sum, to answer the three proposed research questions, this study employs Twitter content analysis to obtain the insights about online discussions during terrorist attacks (RQ1) and in-depth interview to understand Indonesian dual screeners’ media behavior in different stages of the crisis response process (RQ2) and their motivations to use dual screening under the KM bombing crisis (RQ3). The Twitter content analysis examines in total 7,101 tweets while the interview data are gathered from 21 Indonesian dual screeners from three different groups: incident-related actors, digital participant, and journalists.
DOI:10.6814/THE.NCCU.IMICS.006.2018.F05 CHAPTER 4
RESULTS
This chapter outlines the results of Twitter content analysis and interviews. The first section presents the overview of the tweets’ trends, prevalent messages about the KM bombing, and the identification of key mobilizers as well as their tweets’ contents about the KM bombing. The next section describes respondents’ dual screening behaviors during the terrorist attack. It also encompasses the Indonesians’ dual screeners’ media behaviors in different stages of the crisis response process and their motivations to use dual screening during terrorist attacks.
4.1. Twitter content analysis results: Twitter Trends and key messages
In response to RQ1, this study analyzed 7,101 tweets bearing the virally spread hashtag
#KamiTidakTakut which was used to respond to the KM terrorist attack. This section further elaborates the results of Twitter content analysis. After presenting the trends of the tweets from the content analysis and the word cloud analysis, the relations of Twitter content with Google Trend data and offline activities are also discussed in this section. Moreover, this part explains the most prevalent messages appeared on Twitter and analyzes the top 15 influential Twitter mobilizers during KM bombing attack.
4.1.1. Twitter trends of KM bombing attack
When analyzing the trends of tweets during the KM bombing crisis (May 24 to June 5 2018), the period can be divided into two parts: storm (when the Twitter conversations reached the climax) and post-storm (when the conversations started to subside) (see Figure 2). During the
DOI:10.6814/THE.NCCU.IMICS.006.2018.F05 storm phase, the number of tweets increased from 1,994 tweets on May 24 to 4,173 tweets on May 25. The number of tweets on May 24 was lower compared to May 25 as the bombing detonated at 9 pm, leaving only three hours for Twitter users to post their remarks on May 24. The tweets plummeted to 682 tweets on May 26. On the post-storm phase, the tweets declined from 73 on May 27 to only two tweets counted on June 5. Among 7,101 tweets retrieved, there were 5,326 unique users who posted tweets embedding the hashtag #KamiTidakTakut. A total of 4,398 unique users (82.58%) tweeted only once during the period of time. The biggest number of tweet posted by a single user was 28 times.
Figure 2. Distribution
of Tweets during KM Bombing (#KamiTidakTakut)
A word cloud analysis for tweets during the storm phase (May 24-26 2017) and post-storm phase (May 27- June 5 2017) was performed to visualize the most dominant words or themes of tweets during these two phases based on the frequency of words used on the tweets. During the storm phase (see Figure 3), the words appeared were highly related to moral support, not only for the victims, but also for the community (e.g., Jakarta citizens) in general, for instance
Storm Post-Storm
DOI:10.6814/THE.NCCU.IMICS.006.2018.F05
“PrayForJakarta” which reflects offering prayer to Jakarta (the city where the bombs blasted).
Apart from that, there were also “turut”, “berduka”, “cita” (deep condolences), “semoga”
(hopefully/please let), and “safe”. Some other words, such as “bisa” (can), “kita” and “kami” (we),
“mari” (let us), “bersama” (together), “tetap” (still), and “lawan” (counter), spoke about optimism and sense of togetherness after the terrorist attacks.
Figure 3. Word cloud for storm phase
During the post-storm period, the word cloud had different directions. Several new words emerged, indicating new thrust or action-provoking, such as “KamiAkanLawan” (we will fight),
“AksiSimpatikLawanTeroris” (sympathetic act to counter terrorism), “TendangTerorismeISIS”
(kick ISIS terrorism), “GebukRadikalisme” (beat the radicalism), “GebukTeroris” (beat the terrorist), and “waspada” (beware). Interestingly, a lot of words related to the authorities might appear as Twitter users quoted what these authorities said or conveyed their moral support to them.
These words were “Jokowi” (Joko Widodo, Indonesia’s president), “DivHumasPolri”
(Department of public relation, National Police of Indonesia), “Polri” (National Police of Indonesia), “kamihumaspolri” (We are the public relation of National Police of Indonesia),
DOI:10.6814/THE.NCCU.IMICS.006.2018.F05
“Aiptu” and “Hiskam” (Aiptu Lalu Hiskam, Indonesian Police), and “LukmanSaifuddin”
(Minister of Religion).
Figure 4. Word cloud for post storm phase
Figure 5 presents the trend generated by Google Trends, describing the public’s interest in the topic over time. Corresponding with the tweet distribution, the peak of the public’s interest also occurred on the second day (May 25, 2018), plunged dramatically on May 26, and kept declining steadily onward.
Figure 5. The trend of the public’s interest on the KM bombing based on Google Trend
DOI:10.6814/THE.NCCU.IMICS.006.2018.F05 Furthermore, the offline activities regarding the KM bombing attacks were also identified (see Appendix C). The important offline activities mostly occurred during the storm phase (May 24 to May 26). The two suicide bombs detonated on May 24 2017 evening in KM bus station, Jakarta. Not long after the bombs blasted, the national police started to conduct further investigation. The hashtag #KamiTidakTakut (we are not afraid) also exploded on Twitter and became the worldwide trending topic on this social media platform. The following day, on May 25 2017, Indonesia’s president, Joko Widodo made his official speech to respond to the terrorist attack and expressed his condolences. The National Police Chief, Gen. Tito Carnavian, also made his official statement and released the names of the victims. One of the victims was also buried that day. The bus stop was started to be repaired. The next day, May 26 2017, the National Police Chief released the name of the two suicide bombers after prior identification and the national police also arrested five more people suspected to be related to the incident. ISIS also claimed responsibility of the bombing attacks. Furthermore, the bus stop returned the operation regularly.
Considering the aforementioned activities, it might explain why the tweets were dominantly posted during the storm period (May24-26 2017).
4.1.2. Messages types of Twitter communication during terrorist attacks
The results of content analysis from 7,101 tweets indicated that nearly half of total tweets (48.76%) posted action-related content, while emotion-related content and information-related
The results of content analysis from 7,101 tweets indicated that nearly half of total tweets (48.76%) posted action-related content, while emotion-related content and information-related