The Warning, Alert, and Response Network (WARN) Act of 2006 called for the creation of a national all-hazards alerting system that would use multiple technologies to better reach affected populations. Since its passage, there has been an increasing interest in exploring the use of social media to provide alerts and warnings. Social media is a loosely defined term that refers to a set of Internet-based tools that support social interaction through many-to-many communications. The term encompasses a variety of technologies including weblogs, microbloging and mashup tools, and online social networks, examples of which include Twitter, Google Maps, Facebook, and Flickr. It also includes tools like Ushahidi that were purpose-built for use in crises. Social media are being used in all sectors of society to support communication, collaboration, and information collection and dissemination. Social media have proven useful in a crisis both to officials seeking to deliver alerts, warnings, and other information to the public and to citizens communicating with officials and each other.
This report presents a summary of a February 2012 workshop organized by the National Research Council’s (NRC’s) Committee on Public Response to Alerts and Warnings on Using Social Media: Current Knowledge and Research Gaps. The first session of the workshop provided an overview of alerts (an alert indicates that something significant has happened or may happen), warnings (a warning typically follows an alert and provides more detailed information indicating what protective
action should be taken),1 the use of social media for delivering alerts and warnings, and other applications of social media in disaster management. Dennis Mileti, University of Colorado, Boulder, described what is known about how the public responds to alerts and warnings. Kristiana Almeida, American Red Cross (ARC), described how the ARC uses social media during disasters and provided results of ARC research on social media use. Edward Hopkins, Maryland Emergency Management Agency, discussed barriers to the use of social media by emergency managers. Emre Gunduzhan, Johns Hopkins University Applied Physics Laboratory, described current and emerging technologies for disseminating alerts and warnings and enhancing situational awareness using social media.
More than 60 years of interdisciplinary research on disaster response has yielded many insights about how people respond to information indicating that they are at risk and under what circumstances they are most likely to take appropriate protective action. Much of this knowledge has been captured in the “Annotated Bibliography for Public Risk Communication on Warnings for Public Protective Action Response and Public Education” that lists more than 350 publications.2 This body of research covers natural disasters such as Hurricane Camille and the Mount St. Helens eruption; terrorist attacks such as those on the World Trade Center in 1993 and 2001; hazardous material spills such as those that occurred during the 1979 Mississauga, Ontario, train derailment and the 1987 Nanticoke, Pennsylvania, factory fire; building fires such as those at the MGM Grand Hotel in Las Vegas in 1980 and at Chicago’s Cook County Hospital; and technological accidents such as the 1979 incident at the Three Mile Island nuclear power station in Pennsylvania. Mileti outlined some of the key results from research on how the public responds to alerts and warnings:
1 The difference between alerts and warnings can be unclear because a warning can also serve as an alert, and an alert may be accompanied by some information about protective measures. Technology has further eroded the distinction. For example, on mobile devices, the Commercial Mobile Alert service will simultaneously deliver both a distinctive tone (the alert) and a brief message with additional information (a warning). Similarly, sirens have evolved to provide both a siren sound and a spoken message.
2 The extensive “Annotated Bibliography for Public Risk Communication on Warnings for Public Protective Actions Response and Public Education” was compiled by Dennis Mileti, Rachel Bandy, Linda B. Bourque, Aaron Johnson, Megumi Kano, Lori Peck, Jeannette Sutton, and Michele Wood and is available at www.colorado.edu/hazards/publications/informer/infrmr2/pubhazbibann.pdf.
• Research using hypothetical scenarios does not faithfully represent public response. Studies on how people would respond to hypothetical events do not generally predict how the public is likely to behave in response to an actual alert or warning—or during a real event. By contrast, although they may be harder or more expensive to conduct, studies of actual events yield much better insights about the situational determinants of behavior during emergencies.
• Education about the warning system is needed before an event. Public education on warning systems is an important complement to education about how to prepare for disasters. For example, it is important that people know that a certain television or cell phone tone designates an alert or warning, and where additional sources of authoritative information can be found. Furthermore, it is important to remember that the majority of people cannot remember what a given siren means, what a color code may represent, or even the difference between the watches and warnings issued by the National Weather Service. Comprehensive education programs teach the public that a hazard exists, inform them about the alerting/warning systems in place in their communities, and outline what protective actions they might be asked to take. Essentially, these programs prime the public by removing surprises and reducing confusion in a future warning event.
• Alerting needs to attract attention. A primary goal of alerting is to attract the affected population’s attention so that one can then provide information. The most effective alerts are incredibly intrusive, able to be noticed amidst the cacophony of daily life, and, for those who are asleep, literally loud enough to wake them. The more channels, or mechanisms used to disseminate alerts and warnings, the greater the chance an individual will receive a message. Warnings should also be repeated; repetition commands attention, fosters confirmation, and prompts protective action.
• People seek social confirmation of warnings before taking protective action. Before acting in response to a warning, people generally seek confirmation from others. The resulting process is known as milling, in which individuals interact with others to confirm information and develop a view about the risks they face at that moment and their possible responses. Milling creates a lag between the time a warning is received and the time protective action is taken. Yet few approaches to formulating and delivering warnings focus on shortening this milling time. Indeed, one important lesson from past research is that without careful attention to the process of milling, the introduction of new warning systems may increase rather than decrease individuals’ delay in response. Social media provide a new way for these interactions among individuals to occur. Although many first responders believe that social media have given
them less control over the warning process, informal dissemination of messages has always played an important role in the warning process. Indeed, one might conjecture that the inherently social nature of social media might help reduce milling time, but whether this is true and under what conditions is an open research question.
• Messages should contain information that is important to the population. The key content of messages is what (the protective actions that should be taken); when (by what time the protective actions should be taken); where (the geographic area that will be affected); why (the risks and how protective action would reduce their impact); and who (the individuals or entities providing the information). In addition to content, message style also matters. Messages need to be clear, simply worded; specific (i.e., precise and non-ambiguous); accurate (i.e., free from errors that can create confusion); certain (i.e., authoritative and confident); and consistent. When changes in instructions are required, the reasons should be explained.
• Responders should consider the demographics of affected populations when preparing warning messages. Status differences (e.g., gender, sex, age, ethnicity, socioeconomics, and family relationships), an individual’s or a community’s past experiences, and other environmental and social factors all can affect how a warning will be interpreted.
• Access for those with disabilities must be considered when developing alert or warning systems. Individuals with impaired hearing or impaired vision or those with limited abilities may face challenges in receiving particular alerts or warnings. Furthermore, individuals with disabilities may encounter unique challenges in taking protective action or there may be important unique factors in how they may interpret a warning.
• Alerting and warning is a process, not a single act. The communications process includes issuing a warning, monitoring the public’s response to the warning, listening for incorrect information the public may be receiving, and rewarning based on observations of what the public is doing or not doing. Given the mix of official and unofficial information available to the public, it is almost inevitable that people will be exposed to some incorrect information, which can lead to inconsistencies that can delay protective actions. Such misinformation and misinterpretation need to be addressed in subsequent warnings.
With a growing fraction of the public using social media, there has been increasing interest in their use during disasters. One source of information on how social media are used, presented by Kristiana Almeida,
comes from a 2011 American Red Cross study3 in which approximately 2,000 people were interviewed. Nearly 50 percent of those interviewed reported that they visited one or more social media sites almost every day. For those in metropolitan areas, this percentage was higher. Although one in six subjects reported that they had used social media to find information about an emergency, television continues to be their primary source of information during these events. However, in cases where people did not have access to a television, Facebook became their primary information source. Furthermore, more than 80 percent of those interviewed indicated that they were willing to post information about a disaster on social media sites, including images or video clips.
More than a third of the regular social media users interviewed in the Red Cross study indicated that they would request help via social sites, and of those who would post requests for help, 80 percent expected a response within an hour. However, only 15 percent believed that emergency management agencies were actively following social media during emergencies. That is, although many believed that emergency officials should be following their feeds, few believed that they actually were.
Several key lessons could be taken from the 2011 Red Cross study, observed Almeida. First, it suggests that there are opportunities for emergency managers to use social media “crowdsourcing”4 to supplement their situational awareness during emergencies. Second, because (according to the Red Cross study) public officials are not currently prepared to respond to requests for assistance via social media despite growing public expectations for such a capability, officials will need to continue to remind the public to use 911 to call for assistance.
Although some emergency management organizations have begun using social media to interact with the public, several barriers and challenges remain to more widespread adoption, including limited understanding of social media, concerns about loss of control, and institutional limitations. Drawing on lessons he learned from discussions with fellow local emergency managers, Edward Hopkins, Maryland State Emergency Management Agency, noted the follow challenges:
3 American Red Cross. Social Media in Disasters and Emergencies. 2011. Available at http://www.redcross.org/www-files/Documents/pdf/SocialMediainDisasters.pdf.
4Crowdsourcing refers to the use of information provided by multiple, and often many, individuals, who may have volunteered this information or been paid or given some other incentive to provide it. Chapter 3 explores the use of crowdsourcing to win a competition.
• Limited understanding. Despite growing appreciation of the role that social media can play in disasters, many emergency managers remain more comfortable with traditional media, and not all are aware of the potential advantages of social media as a tool for alerts and warnings. Some of this is surely a result of generational or cultural differences. Familiarity and comfort with social media for emergency management can be expected to grow as training opportunities are provided and newly hired employees bring with them a greater familiarity with social media. (The adoption of social media by practitioners is discussed further in Chapter 2.)
• Loss of control. When they use social media, emergency officials cannot control which information social media users share, which raises concerns that they might lose control of messaging or face civil liabilities if misinformation is shared. By contrast, in the traditional command-post style of information dissemination, long-standing relationships between the press and emergency managers provide some sense of control over what information is disseminated as well as well-understood opportunities to disseminate corrections as needed. It is the belief of some officials that with social media, misinformation may spread more rapidly and continue to spread even after a correction is issued.
• Institutional limitations. Especially in an era of shrinking budgets, it is hard to find the resources to evaluate and adopt new tools and technologies, or to invest in the training necessary for their use. Also difficult is securing new resources to cover the additional staff time necessary to monitor social media activity. As a result, although emergency management personnel may try to experiment with social media use, they may find that their other (day-to-day and emergency) responsibilities crowd out the possibility for this work. Also, such experimentation with social media often precedes the creation of guidelines or formal policies for their use, which can lead to unforeseen complications or questions being raised by managers, and implementation of measures that restrict the use of social media.
Emre Gunduzhan discussed technologies that are needed to disseminate public alerts and warnings using social media and to develop situational awareness during disasters using information gleaned from social media. Although some of these tools may be available now, they might not have been used in emergency management practices. Additional tools will also have to be developed that focus on the specific needs of disaster response, commented Gunduzhan, who also outlined the following set of technical challenges common across multiple alerting systems:
• Standard message formats. The Common Alerting Protocol (CAP) is a recently adopted protocol being used across multiple alerting platforms.5 The advantage of CAP is that as an XML-based standard, it is machine-readable, and Web services and applications can receive and process these alerts. More recently, an Internet Engineering Task Force (IETF) working group, Authority-to-Citizen Alerts, has been developing a new protocol for transmitting alerts to the public over the Internet.
• Capabilities for authorizing users. Some sort of scheme, which might be a centralized, federated, or distributed identity management system, is needed to help ensure that only those properly authorized to do so can issue official messages.
• Digital signatures that indicate who sent a message. Public response is affected by who sends a message. The public needs to be able to verify who has sent a message and that the message is authentic. Nonrepudiation supports the assignment of responsibility for the issuance of an alert and protects from potential liability parties who relay an official message in good faith.
• Geotargeting of alerts and warnings. Controlling the granularity of alerting by geographically limiting the area targeted for receiving an alert might greatly protect the public from alert overload and better ensure that information gets to those who need to take a particular protective action (e.g., shelter in place versus evacuate). With respect to this capability, social media services diverge prominently from other alerting systems. In other existing communication channels, physical parameters, such as the area served by a set of cell towers or the metropolitan area served by a radio or television broadcast tower, constrain the distribution of alerts to a specific geographical area. By contrast, social media services embody limited and imperfect knowledge of the precise location of their users. Most often the information available is the address or town that may have been provided as part of a user’s profile; a user’s Internet Protocol address may also provide clues about his or her location. Although users accessing social media services via mobile devices may be able to provide accurate geotargeting information, users generally need to enable this feature, and many choose not to do so owing to privacy concerns.
There are two possible methods for using social media to deliver alerts and warnings. One is for the entity issuing the messages to simply be registered as a user of the service, as would any other information provider. The other is for the entity to establish a special relationship with
5 OASIS. Common Alerting Protocol, v. 1.1. OASIS Standard CAP-V1.1, October 2005. Editors: Elysa Jones and Art Botterell. Available at http://www.oasis-open.org/apps/org/workgroup/emergency/download.php/14205/emergency-CAPv1.1-Committee%20Specification.pdf.
the social media service. The characteristics of each method are outlined in Box 1.1.
Today, emergency managers are making use of only the first alternative, noted Gunduzhan. For example, a local jurisdiction may set up a Twitter account and simply post warnings as an update that would be seen by those who follow that account or find the post as the result of a search. Similar possibilities exist with other social media services like Facebook and Google+, although who sees what messages and under what conditions is more complicated owing to the service design. Such arrangements have the advantage of being possible to implement today without reaching agreement on a new interface to deliver alerts and warnings or new arrangements for their display to users at risk from an event. On the other hand, they require that emergency managers find a set of clients that allow them to post messages to each service, and no special provisions are available to ensure that affected populations see the right set of alerts and warnings. (There is also a question of how to establish the authenticity of alerts and warnings; see above.)
Alternatively, emergency managers might establish partnerships with social media services for disseminating alerts and warnings, either on behalf of individual agencies or collectively at the state or federal level.
Characteristics of Alternative Methods for Distribution of Social Media Alerts
Distribution by a Registered User of Social Media Services
• Alerts appear as ordinary messages.
• Alerts are difficult to scale and maintain due to many different application programming interfaces (APIs) that are subject to change at any time.
• Citizens have to register for the service (opt in).
• Alert authentication and authorization are managed externally to the social media site.
• Non-repudiation is not a major issue.
Distribution Via Collaboration with Social Media Sites
• Alerts can be given priority and special treatment.
• A single standard interface definition can be used by all collaborating social media sites.
• Citizens will automatically be included in the service but may be allowed to opt out.
• The social media site implements additional alert authentication and authorization.
• Non-repudiation and liability protection need to be addressed.
For example, social media providers could agree with emergency managers to take steps in concert to make alerts more obvious to users. Such an approach is analogous to the Commercial Mobile Alert Service, in which cellular carriers have been working with the federal government to establish a standard message format, alert tone, and national gateway for delivering messages.
Several workshop participants observed that unlike other means of delivering alerts and warnings, social media technologies offer the distinct advantage that they are two-way, are interactive, and provide an opportunity to see how the public is responding to the message. That same interactivity poses new challenges, however, such as managing the sheer volume of messages that can be sent during a major event. For example, following the announcement of Osama Bin Laden’s death, messages (“tweets”) were being posted to Twitter at a rate of about 3,000 per second.
Given the flood of information available from social media, some workshop participants observed that automated filtering tools are likely to be key to developing situational awareness. For example, automation can help categorize sources and separate relevant and irrelevant information. Progress toward such automated tools will rely on advances in natural-language processing. The location of a source of information is also helpful in separating out relevant information and can sometimes be determined from a user’s profile or from metadata associated with the message, or inferred from the content of the message itself. Several workshop participants noted that even with greater automation, human judgment will be needed to interpret and act on the selected messages.6
Visualization tools that can help emergency managers make sense of social media information are already available to some extent, observed workshop participants. Ushahidi, an open source tool for information collection, visualization, and interactive mapping,7 for example, has been used successfully during a number of disasters.8 Looking ahead, one
6 In an effort to combine technological tools and human judgment to monitor social media, the American Red Cross opened its Digital Operations Center in March 2012.
8 The value of Ushahidi during the 2010 earthquake in Haiti is explored in Nathan Morrow, N. Mock, A. Papendieck, and N. Kocmich, Independent Evaluation of the Ushahidi Haiti Project, Development Information Systems International, 2011, available at http://www.alnap.org/pool/files/1282.pdf. The report explains, “Perhaps the most common use of information aggregated by UHP was for situational awareness. The Department of State
challenge is how to integrate information from multiple social media services as well as from other information sources. Gunduzhan noted that addressing this challenge will require attention to a mix of standard interfaces and formats as well as translators.
The use of tools to gather and analyze information derived from social media services raises a set of privacy issues associated with the collection, processing, retention, and distribution of such information. These issues were explored in another workshop panel, whose observations are presented in Chapter 5.
During the discussion, panelists and other workshop participants offered a number of observations on the use of social media for alerts and warnings, including the following:
• Although the public response to alerts and warnings has been studied for some time, as have more general questions about the information needs of the public during disasters, there has been comparatively little research on the use of newer technologies such as mobile devices (e.g., cell and smart phones) or social media during disasters. Research would help shed light on such key questions as how the new technologies could help shorten the milling time between receipt of an alert or warning and the taking of protective action.
• Questions still remain regarding the extent to which social media represent a truly new source of information for improving the situational awareness of emergency managers or whether much or most of the information simply repeats information already available from other, traditional sources.
• The particular characteristics of social media platforms yield different “affordances” of use during disaster situations. For example, social networks on Facebook are usually user-defined: messages to one’s network might help target localized attention and allow extended discussion. Facebook newsgroups, however, are public and can draw collected attention. Twitter supports rapid communications that are most often public. The communications are short and can easily be “retweeted” or
analysts for the USG interagency task force used Ushahidi in at least one case to help triangulate conclusions about the situation on the ground, and US military organizations used Ushahidi data feeds along with other sources in a similar manner to inform their early situational assessments. There is also some evidence of the information being used for specific operational and tactical actions targeting specific communities (and to a much lesser extent, individuals). US marines used the information to identify ‘centers of gravity’ for deployment of field teams to [affected] areas.”
propagated publicly. Those that are propagated have a chance of receiving attention; those that are not die out quickly.
• It is important not to think of these platforms as better or worse; rather, it is critical to understand that they are different places along sometimes circuitous paths to seeking and finding information.