A new study examines tweets from various disasters, providing key information for emergency managers and their communications teams.
Throughout the last few years social media has become a key communications strategy for emergency managers. Whether it’s for sharing preparedness messages during blue-sky times or getting crucial information out in real time during an emergency, platforms like Twitter and Facebook are now part of nearly every agency's public-outreach plan. This evolution in crisis communications has been followed by many, and a recently released study sought to understand what affected populations, response agencies and other stakeholders can expect from tweets in various types of disaster situations.
The study, What to Expect When the Unexpected Happens: Social Media Communications Across Crises (PDF), examined tweets posted during 26 emergency situations in 2012 and 2013. With the goal of measuring the prevalence of different types of tweets during various situations, the researchers examined both the information and its source.
The tweets were classified into six categories, and researchers determined the average percentage of tweets for each: affected individuals (20 percent), infrastructure and utilities (7 percent), donations and volunteering (10 percent), caution and advice (10 percent), sympathy and emotional support (20 percent), and other useful information (32 percent). Tweets classified as other useful information varied significantly, the report says. “For instance, in the Boston bombings and LA Airport shootings in 2013, there are updates about the investigation and suspects; in the West, Texas, explosion and the Spain train crash, we find details about the accidents and the follow-up inquiry; in earthquakes, we find seismological details.”
The tweets were also categorized by source: traditional and Internet media accounted for 42 percent on average; 9 percent were from eyewitnesses; 5 percent were from government officials and agencies; nongovernmental organizations posted 4 percent of tweets; 2 percent were from businesses; and 38 percent came from outsiders, which the researchers wrote can vary greatly depending on the event.
The relatively low percentage of tweets from public-sector sources “is because governments must verify information before they broadcast it, which takes considerable time. … The [seven] events with the highest percentage of tweets from governmental agencies are due to natural hazards, progressive and diffused (7 to 13 percent), which are the cases when the governments intervene to issue or lift warnings or alerts.”
The researchers — Alexandra Olteanu of Ecole Polytechnique Federale de Lausanne in Switzerland, and Sarah Vieweg and Carlos Castillo, both with the Qatar Computing Research Institute — replied to a series of questions from Emergency Management via email. Their responses are below.
What was your goal with the study and what were you hoping to learn?
Our primary goal was to measure the prevalence of various types of information found in Twitter messages broadcast during different types of crisis situations. We want to provide an indication of what kind of information (and from whom) response agencies and other stakeholders can expect — and not expect — to find on Twitter. Specifically we looked for similarities and differences in Twitter communications that happened across different crisis events according to the various characteristics of said events.
How can emergency managers and responders (and their communications teams) use this information?
The takeaways here are twofold, depending on their role of either consumers or producers of social media content during crises. Consumers of social media content must consider that to identify relevant information that is not present in mainstream media or Internet media sources, they will need to skip through approximately 80 percent of crisis-related tweets on average. That number jumps to more than 90 percent if they are interested in eyewitness accounts. Similarly if they are interested in locating only information about particular aspects of a crisis (e.g., affected individuals, caution and advice, infrastructure and utilities, etc.) then more than half of the tweets will need to be discarded on average. In fact, these are lower bounds; often not all the messages of a given type will be useful. Noise is a considerable problem, as is locating specific information. Additionally, those who develop and maintain social media tools for emergency response should consider that while there are broad classes of information that are likely to be prevalent on Twitter independently of the type of crisis, a significant portion of messages will depend on the specificities of each crisis.
As content producers, we found that only 4 percent and 5 percent (on average) of the information available on Twitter originates from governmental and nongovernmental agencies (including cases when the information is relayed by other Twitter users not affiliated with such agencies). In addition, during crises that are preceded by a warning period (typically those caused by natural hazards), government messages appear early. In instantaneous crises that do not allow for pre-disaster mobilization, such messages tend to appear relatively late with respect to other sources. To mitigate this gap, they might consider the use of standardized hashtags (similar to a policy already employed by the Filipino government) — a solution extensively covered by our colleague Patrick Meier.
Was there anything that stood out to you or was the most surprising finding?
Every crisis generates a unique reaction — sometimes the most common type of messages in one crisis (e.g., eyewitness accounts in the Singapore haze crisis in 2013) will be absent in another (e.g., eyewitness accounts in the Savar building collapse in 2013). Yet, when we looked at the Twitter data at a meta-level, we noticed similarities in terms of types of information users tend to post given a particular dimension of the crisis: Human-induced crises tend to be more similar to each other than to crises induced by natural hazards. Similarly, instantaneous crises are also more similar to each other than to crises that develop gradually. For example, messages about caution and advice from governmental sources are less common in instantaneous ones, while messages about donations and volunteering were less prevalent in all human-induced crises (10 percent or less).
How did you choose which disasters to study?
We started with Wikipedia’s event list (for instance, for January 2013 we consulted http://en.wikipedia.org/wiki/January_2013), from which we then chose events with at least 100,000 tweets associated with them. While doing this selection, we also tried to cover as many crisis types as possible.
Is there anything you’d like to add for Emergency Management’s readers?
Social media use in disaster situations is here to stay, and it is important for those in emergency management to not only learn how members of the public are using it to communicate, gather and disperse information, but to know how they can leverage social media to their greatest benefit. To effectively make use of social media in general — and Twitter in particular — one needs to know the kind of information one can expect to find during different types of crisis situations.