Micro-blogging lets the public stay involved in risk communication following disasters. Here, we explore the patterns of risk communication on Twitter regarding the nuclear radiation threat in the aftermath of the 2011 Fukushima earthquake and tsunami, focusing on patterns of retweets of alarm and reassurance and providing insight into micro-blogging behavior and its consequences.
Key Insights
- Risk communication can be viewed through the lens of the Twitterverse, particularly in the form of retweets, as reflected in the Japanese public’s fears over the 2011 Fukushima radiation emergency and the Japanese government’s related effort to calm them.
- In the days following the disaster, the public was more concerned about the dead and missing and the tsunami’s devastation, but fear over the safety of the affected nuclear power plant quickly came to dominate.
- While government sources produced more reassuring tweets than did ordinary citizens, they were eventually retweeted less, signifying their loss of influence.
Communication is important in emergency response.5,10,11,15,16,18 In the aftermath of the 2011 Fukushima earthquake and tsunami, Twitter was an important social medium for distributing information to millions of people worldwide, including in Japan, with the Japanese government sharing emergency information, relief organizations sharing shelter information, and ordinary citizens posting news of their local situations.25
Radiation fears were of utmost importance to the Japanese people. However, the Japanese government and Tokyo Electric Power Company (TEPCO), owner of the crippled nuclear plant in Fukushima, had trouble communicating with them regarding the related risk. Moreover, available information included conflicting and contradictory statements and claims. For example, Greenpeace said data from its own scientists largely correlated with official Japanese data,8 while Japanese public broadcasting outlet NHK reported “The Japanese government withheld the release of data showing that levels of radiation more than 18 miles (30 kilometers) from the crippled Fukushima nuclear plant exceeded safe levels.”14 The Japanese public’s response was to begin asking whether the government was indeed telling the truth or perhaps covering up potential risks that could prompt public panic.
Pew Research reported that Twitter included more than 500 million users worldwide as of February 2012 (http://www.pewinternet.org). Among U.S. Internet users, 15% used the service, with 8% doing so on a daily basis. A common Twitter behavior called “retweeting” involves users passing on or sharing the tweets they find of interest to their own followers and contacts. Among the millions of messages disseminated each day, retweets reflect the information that most strongly affects Twitter readers. Retweeting implies at least one reader viewed the information as important enough to want to share it, making retweets more reflective of and influential on the mood of the general population, which hears from a much greater number of observers than it would through a typical lone, unrepeated Twitter message (http://www.retweetrank.com). Our focus here is on how retweets were used to spread both alarming and reassurance information.
Research Background
Use of social media by any government limits the lag in information flow from traditional mass media to individuals and from individuals to other individuals in the form of models (such as the “two-step flow” of communication).9 Social media offers the government direct access to individuals and other private entities, letting them quickly transfer information in risk communication and information management, disseminating messages of assurance and comfort to the victims of disasters. Risk communication is the focus of information management immediately after a disaster. Certain disaster situations result in prolonged periods of danger and instability as organizations and governments decide how to balance the level of information they should provide, addressing concern over causing panic, and, in some instances, their own reputations and political need to save face. The nuclear radiation danger in Japan in 2011 was one such instance.7 What, when, and how to communicate to the public sums up the concept of risk communication.
“Risk communication has been defined as an interactive process of an exchange of information involving multiple messages about the nature of risk,” according to Chartier and Gabler.4 News and social media play an important role not only providing information but also bringing the public’s attention to urgent matters. However, when experiencing a stressful situation, the public might view risk-related information issued by official sources (such as a government) as biased, incomplete, or incorrect, tending to leave them feeling dissatisfied.4
Reassurance. Many major international news outlets reported the Japanese government withheld release of accurate radiation data,13 possibly to avoid alarming the public, as it tried to provide reassurance. Prior research explored reassurance, though not in the context of micro-blogging; for example, Barnett et al.1 weighed the public perception of precautionary advice originating from the U.K. government in 2007 on possible health risks from mobile phones, observing whether risk was presented as less serious than it truly was, leading to public anxiety, panic, or anger when the truth was ultimately revealed. Other times it can be important for a government to attempt to purposely get people concerned. An important matter to understand is how the Twitter environment fosters a balance between alarm and reassurance and how the public receives messages from the government under various circumstances. Ungar24 explored media reassurance under “hot crisis” conditions, trying to identify the scenarios in which reassuring coverage is more likely than alarming coverage, and found reassurance much more likely in the immediate aftermath of a disaster when panic can spread quickly.
Analysis
We emphasize how retweets echo through the Twitterverse to determine whether the information shared in the aftermath of the Japanese earthquake and tsunami produced reassurance or alarm. Emphasizing the tone of the retweets, we explore what information the public shared. We also explore how the Japanese government communicated through Twitter and how the information was received in the context of reassurance despite the risky situation. Weighing Twitter use in this context will improve how it is managed by governments, as well as by the public, in future disasters.
Data collection. Using the Twitter search API with a service called “Twapper Keeper,” we collected several hundred thousand tweets containing the term “Japan radiation” in the month following the earthquake. “RT @” and “via username” were the two most common ways we used to distinguish retweets from regular tweets, helping us narrow the dataset to 38,300 retweets regarding radiation in the month following the earthquake. While it was possible that retrieving tweets based on “Japan radiation” might not have been about 2011, the fact that the data was collected at the time of the incident gave us confidence as to the accuracy of the sample. We did not include other alternative conventions for retweets in our investigation. We also did not clean the data to account for bots, though our inspection of the most frequently cited users in our raw data found none were retweeting and thus not included. We identified and focused on only the top retweeted messages.
Coding schema. To determine the best way to measure characteristics of alarm and reassurance of our included retweets, we examined the previous literature; for example, Stephens and Edison23 used a simple classification “measuring the number of reassuring or positive statements versus alarming or negative statements,” and Speckens et al.20 developed questionnaires for measuring reassurance (in the context of whether patients feel reassured by their physicians). We adapted some of it to our own research context to develop our coding schema; for instance, reassurance and alarm are not the only dimensions a retweet can have. Likewise, a tweet not expressing alarm is not by default reassuring and vice versa. Following the literature, we created a codebook to help determine if the author of a tweet was communicating alarm, fear, or something else. Coding categories and associated questions included:
Alarm. Did the tweet communicate worry and indicate the situation is dangerous?; and
Reassurance. Did the tweet communicate fear (reverse coded) or communicate calm?
We also measured another dimension:
Doubt. Did the message communicate doubt about the situation?
Finally, we were also interested in whether a particular tweet was from the Japanese government, so we coded two additional questions: Was the message from (or related to) government?, and if it was, did it represent a direct quote or a secondary quote?
Inter-Coder Reliability
We hired two graduate students to work 100 hours each. Both had prior experience on projects involving the coding of tweets. Before the formal coding process began, we conducted a pilot coding session with a randomly chosen 50 tweets. The Kappa value of .90 was high, confirming coding reliability.
Descriptive analysis. Figure 1 outlines the distribution of the retweets for the month following the 2011 earthquake and tsunami. We observed how the retweets and their origin changed over time as the public’s concern unfolded. Public interest in the nuclear question began moderately but shifted over time. The public was more focused on the dead and missing and the raw devastation of the tsunami. Meanwhile, the possibility of a nuclear meltdown began to emerge. Indeed, there was still the expectation that cooling of the nuclear plant would resume on March 12 and therefore radiation was a relatively trivial issue in the larger picture of the earthquake and tsunami.3 However, public fear for the safety of the nuclear plant continued to grow as TEPCO failed to bring the situation under control. The number of retweets with the words “Japan radiation” increased steadily as the situation evolved. Twitter retweet activity peaked in the final days of March as the possibility of a meltdown continued and the public’s radiation fear increased dramatically.26
Reflecting evolving risk communication, the table here lists the top 25 most commonly retweeted messages and their frequency in the month following the earthquake and tsunami. Messages originating with the government are highlighted in green; others are in blue. Nine conveyed information from a government agency, or 30% of the most frequently retweeted content. The rest was of a more independent nature communicated by a variety of sources. Of the 25 messages, with the exception of two from Twitter users, 23 had all been passed on from major traditional media.
Among the top 25, only three took a reassuring tone, with most reflecting alarm or caution. The top messages sought to avoid overassurance, including “Low levels of radiation found in U.S. milk”; “Japan’s chief cabinet secretary says it could be several months before radiation stops leaking from Fukushima nuclear plant”; and “Radiation in water rushing into sea tests millions of times over limit.” All managed to only increase public concern. Looking further at the top 25 retweets, at least 12 conveyed alarm and five reassurance. Given the information regarding radiation levels and concern expressed in the top retweeted messages, the general public had messages of both alarm and reassurance, reflecting both sides of the picture.
Results
Of the 38,300 retweeted messages in our sample, our two coders looked at 50 random messages to get a sense of the tone of the messages; we then randomly selected 1,520 more through an Excel rand() function. The distribution of the distinct retweets in these messages showed a pattern similar to our full universe of tweets based on a standard data-reduction technique used in large qualitative datasets.12
The coders then independently coded the full sample dataset of 1,520 messages. The Kappa value for inter-coder reliability of the full sample dataset was .977; Kappa value higher than .70 is viewed as acceptable, rendering our analysis rigorous since most other such efforts involve inter-coder reliability for a subsample smaller than ours. Note “distinct retweets” refers to this total pool of 1,520 messages, weighing each message equally. “Total retweets” means the number of times each message was retweeted, or the 9,545 times the distinct messages were retweeted in total.
Our coders coded the questions independently; a positive response to one question did not necessarily require a negative response to another. This way we ensured we were measuring truly different things. We conducted a non-parametric T test for the dimensions. The results (t value = 6.78; prob < 0.0001) indicated a statistically significant difference between the mean values for retweets expressing alarm and retweets expressing reassurance.
The number of tweets concerning nuclear radiation began at a relatively low level, shifting over time. Retweet activity peaked at the end of March when fear of a nuclear meltdown was widespread, after which it decreased gradually through early April. We separated the data sample into three periods reflecting these patterns. The first, period 1, included retweets (579) from before March 29; the second, period 2, included those (493) from March 30 to April 3; and period 3 included retweets (450) from April 4 to April 8. Despite each including a similar number of retweets, each also represents a distinct time period with regard to the disaster and within the micro-blogging community. In the days following the earthquake, little was changing, and the public had not yet focused on the nuclear situation. Twitter activity surged in period 2 with regard to radiation, as a nuclear meltdown became a distinct possibility. Period 3 followed this surge in interest and activity.
Alarm. We conducted nonparametric analysis—Friedman’s non-parametric test in SAS—to determine if the three subsample sets reflected different patterns over time communicating messages of alarm; results included Chi-Square = 28.9036 with p<0.0001, meaning the sub-datasets showed significant differences in the communication patterns conveying messages of alarm. These results also reflect local Fukushima circumstances as the situation unfolded. The great surge in Twitter activity coincided with a notable increase in alarm. Looking at the average of the two alarm questions in period 1, we see these tweets were unamplified, representing 43% of the total number of messages but 47% of the pool of distinct retweets, an amplification factor of 0.91, or 43%/47%. As alarm increased, the Twitter community began retweeting messages expressing alarm more frequently than those expressing lack of concern, thus amplifying content involving alarm. We calculated an amplification factor of 1.19 in period 2 (64%/54%) and 1.32 in period 3 (62%/47%) (see Figure 2).
As the Fukushima situation grew worse, amplification increased greatly. As alarm information decreased, the public remained in a state of alarm, with its total proportion of retweets elevated at nearly the same level. Although there were fewer messages of alarm for them to retweet, they continued to do so.
Reassurance. We performed non-parametric analysis, with results—Friedman’s Chi-square=14.8925, p=0.0019—showing “reassuring” communication patterns differed over the three periods.
As the situation continued to worsen, the public retweeted reassuring content less frequently. During period 2, at the end of March, as fear of a meltdown peaked, messages communicating reassurance decreased substantially. A similar level of low reassurance remained through period 3, reflecting that the Japanese public was on edge as the possibility of a meltdown continued despite more reassuring information from the government and media.
The content of the retweets showed a similar level of reassurance. However, the total portion of public retweets dropped in periods 2 and 3, reflecting non-amplification of the information. The most frequently retweeted messages expressed alarm, while messages expressing reassurance were retweeted less frequently. The public had stopped echoing the reassuring information—the exact opposite of what was seen in the alarm dimension when the community was quick to echo alarm content.
All three periods showed significantly more alarm retweets in terms of number of distinct retweets and number of retweets. Moreover, there appeared to be an inverse relationship between proportion of retweets of alarm and of reassurance.
Messages from the government. In period 1, prior to the surge in Twitter activity, the government’s messages enjoyed significant amplification from the Twitter community. Despite accounting for only 5.7% of the messages being retweeted, the number of times the messages were retweeted accounted for over 14% of the total retweets in our sample of 1,520 retweeted messages, resulting in an amplification factor of 2.5 (14%/5.7%). However, this amplification vanished completely in period 2. The percentage of distinct retweets and percentage of the total retweets from the government were identical at 7.4%, an amplification factor of 1 (see Figure 3).
The voice of the government was either drowned out or ignored by the Twitter community, so, at least in terms of Twitter, had become less influential. Messages originating with the government represented a slightly growing percent of the sample of distinct messages retweeted over time, though they were retweeted proportionally less frequently as interest intensified. The government’s aim was to get more information to the public, but this did not translate into information being distributed more widely. Moreover, the public did not echo the additional information from the government. Amplification of government information partially returned in period 3 when the government was the source of 9.2% of distinct retweets and 13.4% of total retweets. This 46% amplification was far short of the 145% amplification of government tweets in period 1.
As public doubt increased, the public was less likely to amplify the government’s information by repeating it.
Another question we asked was whether a particular retweet communicated doubt. We observed a significant increase in doubt in period 2 that then decreased in period 3. The decline in amplification of governmental information directly corresponded to this increase in doubt. We theorized that as public doubt increased, the public was less likely to amplify the government’s information by repeating it. We drew a similar conclusion regarding how direct quotation of government sources dropped significantly as doubt increased.
Government tweets. While the government had no control over the frequency of its words that were retweeted, it was able to control the amount and content of information it provided. In period 1, 47% of government messages communicated calm, while only 14% of non-government retweeted messages communicated calm (see Figure 4). In period 2, 36% of government messages communicated calm, though it was only 15% of the non-government sample.
Discussion
This analysis reflects strong evidence that the alarming and reassuring tone of the micro-blogging environment changes dramatically as a situation and the public’s mind-set evolve. We also found the information communicated through retweeted Twitter messages can be notably different from that identified in previous research of television and newspaper coverage;23 while that research found traditional media coverage was more reassuring, the micro-blogging universe generally expressed more alarm.
Nevertheless, we noticed a large number of the most retweeted messages originated from traditional media sources, showing how traditional media have embraced this new channel. The effect of the traditional media’s voice is reflected in the high frequency of their content in retweeted data.
We found retweets from governmental sources reflected a much more reassuring message than those not coming from the government, especially in the immediate aftermath of the earthquake (and nuclear disaster) when fear of a meltdown was pervasive among the general public. In order to calm potential panic, government agencies and sources may sometimes look to withhold or postpone releasing negative data, especially when their own reputations and face saving are involved. Over time, government feels less of a need to reassure, shifting toward a more alarmed posture. While governments initially wish to avert panic, they ultimately need to keep the public alert and focused on possible danger.
Conclusion
Our results point to the need for reassuring messages via social media, as well as their value in risk communication, as the micro-blogging universe takes a more alarmed tone compared to traditional media. Subsequent disasters (such as Hurricane Sandy in 2012 on the East Coast of the U.S.) show such reassurance can come not only from the government but also from the private sector; for example, in the aftermath of Sandy in New York City and along the New Jersey coast, Con Edison, the regional electric utility, sent reassuring Twitter messages regarding return of power on particular streets and buildings and estimates of when power would be restored to the rest.22 It did a good job engaging the general public, reassuring it and showing how social media can play a role as important as or more important than word of mouth. However, tweets and retweets are extremely valuable when the major communication channels are down or difficult to access. Public-private partnerships are needed to rebuild communities ruptured during such situations, with social media playing an especially important role.
In this article, we have focused on retweets rather than whether news was broken first on Twitter. The former has received limited attention from the perspective of disaster preparedness and remains a subject for future exploration. Moreover, Twitter is unlike other social media in that it restricts a message to 140 characters; how this inhibits or fosters the exchange of information during emergencies remains unclear and is thus likewise a subject for future exploration. Another issue not addressed here is motivation; for example, why did individuals retweet certain news stories more than others? The secondary nature of our analysis restricted examination of this question because it requires polling retweeters for such motivations and is also likewise a subject for future exploration.
Our current research focused solely on retweets on Twitter rather than on other social media platforms. Perhaps other platforms (such as Facebook) also experienced similar (or different) communication behaviors. Without data either way, our study was limited to Twitter users and followers. Moreover, because it did not focus on the effectiveness of retweets, we did not code the popularity of respective Twitter messages. Hence, a retweet from a Twitter user with many followers might perhaps be more effective than one from one with fewer followers. Finally, we did not know or code for the geographical information of the retweeters. It would be interesting to see how retweets from different geographical locations differ with respect to the geographical epicenter of a disaster, representing limitations of our study and topics for potential future research.
Acknowledgment
This research was funded in part by the National Science Foundation under grants 0916612 and 1134853. The research of the third (corresponding) author, H. Raghav Rao, was supported in part by Sogang Business School’s World Class University Project (R31-20002) funded by the Korea Research Foundation and by the Sogang University Research Fund. Any opinions, findings, and conclusions or recommendations expressed here are those of the researchers and do not necessarily reflect the views of the National Science Foundation; NSF has not approved or endorsed this content.
Join the Discussion (0)
Become a Member or Sign In to Post a Comment