incorporating social media in community emergency …
TRANSCRIPT
The Pennsylvania State University
The Graduate School
College of Information Sciences and Technology
INCORPORATING SOCIAL MEDIA IN COMMUNITY EMERGENCY RESPONSE
A Dissertation in
Informatics
by
Rob Grace
© 2019 Rob Grace
Submitted in Partial Fulfillment
of the Requirements
for the Degree of
Doctor of Philosophy,
May 2019
ii
The dissertation of Rob Grace was reviewed and approved* by the following:
Frederico Fonseca
Associate Professor of Information Sciences and Technology
Dissertation Advisor
Chair of Committee
Andrea Tapia
Associate Professor of Information Sciences and Technology
Jack Carroll
Distinguished Professor of Information Sciences and Technology
Clio Andris
Assistant Professor of Geography
Jessica Kropczynski
Special Member
Assistant Professor of Information Technology
University of Cincinnati
Mary Beth Rosson
Professor of Information Sciences and Technology
Associate Dean for Graduate and Undergraduate Studies
Director of Graduate Programs
*Signatures are on file in the Graduate School
iii
ABSTRACT
The smart city envisions civic officials managing public services through the use of
sophisticated analytics to collect and process data from an array of physical and social sensors.
Included in this vision are emergency dispatchers who intelligently aggregate traditional 911 calls
with “calls” from citizens sending text messages, images, streaming video, as well as social media
posts gathered during an emergency. Accordingly, 911 dispatch centers, known as Public-Safety
Answering Points (PSAPs), transform from reactive call centers to proactive data analytics and
coordination hubs providing first responders with real-time information during, and potentially
before, an emergency.
However, the transformation of PSAPs around new data sources such as social media
requires effective methods to collect social media data in a geographic community and incorporate
social media analysts and analysis tools within emergency dispatch work. In three phases of
investigation, this dissertation i) conducts scenario-based interviews with emergency responders to
outline community emergency response as a design context for social media distribution and
monitoring; ii) introduces and evaluates novel methods to identify hyperlocal social media users
and collect hyperlocal social media data; and iii) articulates sociotechnical requirements for
emergency dispatch work involving 911 call takers, dispatchers, and social media analysts
gathering, analyzing, and synthesizing information from social media users and 911 callers during
simulated emergency scenarios.
Together, the three phases of investigation specify sociotechnical requirements for PSAPs
to collect, analyze, and make use of social media data in ways that can provide early warning of
emergencies and improve situational awareness among emergency responders and citizens.
Moreover, by explicating the sensemaking process in which social media data becomes actionable
in community emergency response, and facilitating co-design with emergency responders through
the use of low and high-fidelity design enactments, this dissertation contributes to theoretical and
methodological approaches to the design of social data analytics for emergency response and
management.
iv
TABLE OF CONTENTS
List of Figures ......................................................................................................................... vii
List of Tables ........................................................................................................................... viii
Acknowledgements .................................................................................................................. ix
Chapter 1 Introduction ............................................................................................................. 1
Dissertation Overview ...................................................................................................... 8 Phase I: Context ....................................................................................................... 11 Phase II: Awareness ................................................................................................. 15 Phase III: Integration ................................................................................................ 18
Outline of Chapters .......................................................................................................... 20
Chapter 2 Community Emergency Response .......................................................................... 24
Distinguishing Emergencies, Disasters, and Catastrophes ............................................... 25 Emergency Management Cycle ....................................................................................... 27 Whole Community Approach .......................................................................................... 30 Community Emergency Response ................................................................................... 32
Emergency Reporting and Dispatch ......................................................................... 33
Chapter 3 Phase I: Context ...................................................................................................... 36
Literature Review: Coordinating Social Media Use During Crises ................................. 36 Information Disseminated on Social Media During Crises ...................................... 38 Using Information Disseminated on Social Media During Crises ........................... 39 Coordinating Social Media Use During Disasters ................................................... 41 Barriers to Social Media Use in Community Emergency Response ........................ 44
Social Media Use as Cooperative and Coordinative Work .............................................. 51 Research Questions .................................................................................................. 54
Method: Scenario-Based Interviews ................................................................................ 55 Scenario-based Interview Design: Scenario Construction ....................................... 56 Participant Recruitment ............................................................................................ 59 Interview Protocol and Data Collection ................................................................... 61 Scenario-based Interview Data Analysis .................................................................. 62 Evaluative Criteria for Qualitative Research ............................................................ 64
Findings: Coordinating Social Media Use in Community Emergency Response ............ 67 Social Media Distribution ........................................................................................ 68 Social Media Monitoring ......................................................................................... 75
Discussion ........................................................................................................................ 83 The Right Staff for the Right Task ........................................................................... 84 Tools for Information Access ................................................................................... 86 Trust in Protocol ....................................................................................................... 88
v
Limitations and Future Work ........................................................................................... 90 Conclusion ....................................................................................................................... 92
Chapter 4 Phase II: Awareness ................................................................................................ 95
Identifying Hyperlocal Social Media Users and Information Sources ............................. 97 Local Social Media Users ........................................................................................ 98 Why is identifying locals important? ....................................................................... 102 How have locals been previously identified? ........................................................... 103 Research Questions .................................................................................................. 109
Collecting Hyperlocal Social Media Data ........................................................................ 109 Deploying Network Filtering ................................................................................... 113 Research Questions .................................................................................................. 114
Method: Social Triangulation and Network Filtering ...................................................... 114 Social Triangulation ................................................................................................. 114 Location, Keyword, and Network Filtering ............................................................. 119
Findings: Identifying Hyperlocal Social Media Users and Information Sources ............ 125 Community Organization Following among Users .................................................. 125 Evaluation of Users’ Locations ................................................................................ 128 Analysis of Local Information Curation Behaviors ................................................. 130
Findings: Collecting Hyperlocal Social Media Data ........................................................ 133 Discussion ........................................................................................................................ 140
Identifying Hyperlocal Social Media Users and Information Sources ..................... 141 Collecting Hyperlocal Social Media Data ................................................................ 150
Limitations and Future Work ........................................................................................... 152 Conclusion ....................................................................................................................... 154
Chapter 5 Phase III: Integration ............................................................................................... 157
Literature Review: Evolution of Next-Generation Emergency Dispatch ........................ 157 Incorporating Social Media in Emergency Dispatch................................................ 160
Theory: From Situational Awareness to Sensemaking .................................................... 161 From Sensemaking to Distributed Sensemaking ...................................................... 164
Methods: Role Play & Simulation ................................................................................... 167 Emergency Dispatch Role Plays .............................................................................. 167 Emergency Dispatch Simulations ............................................................................ 168
Findings: Role Plays ........................................................................................................ 171 “The caller is unresponsive at this time” .................................................................. 172 “Don’t forget to press the period” ............................................................................ 175 “We out here in Bayside” ......................................................................................... 176
Findings: Simulation Role Plays ...................................................................................... 178 Scenario One: Mall Shooting ................................................................................... 178 Scenario Two: Severe Flooding ............................................................................... 183
Discussion ........................................................................................................................ 188 Emergency Dispatch as Distributed Sensemaking ................................................... 189 Design Requirements for Incorporating Analysts in Emergency Dispatch .............. 194
Conclusion ....................................................................................................................... 199
vi
Chapter 6 Conclusion ............................................................................................................... 201
Implications for Community Emergency Response ......................................................... 203 Implications for Theorizing “Actionability” .................................................................... 208 Implications for Crisis Informatics Design Research ...................................................... 213 Directions for Future Research ........................................................................................ 218
References ................................................................................................................................ 221
Appendices ............................................................................................................................... 244
Appendix A Interview Protocol ...................................................................................... 244 Appendix B IRB Approval.............................................................................................. 247 Appendix C Funding ....................................................................................................... 248
vii
LIST OF FIGURES
Figure 3-1. Social media distribution information flow from officials to citizens. ................. 69
Figure 3-2. Social media monitoring information flow from citizens to officials. .................. 77
Figure 4-1. Twitter users following community organizations by asset category. .................. 117
Figure 4-2. Number of community organizations followed among Twitter users (colored
by follower level). ............................................................................................................ 126
Figure 4-3. Categories of community organizations followed among Twitter users (by
follower level). ................................................................................................................. 127
Figure 4-4. Geographic locations of users following community organizations. .................... 129
Figure 4-5. Sociogram of 1-mode affiliation matrix of local organizations (colored by
organizational categories). ............................................................................................... 131
Figure 4-6. Attribute-based model of structuration of community assets (labels sized by
number of followers; ties weighted by co-followership. Upper left shows structuration
of assets among citizens following 2 organizations; upper right shows 3 to 9; Lower
left shows 10 to 49; Lower right shows 50 and over). ..................................................... 132
Figure 4-7. Total on-topic tweets (blue) and situational report tweets (red) in the Location,
Keyword, and Network Datasets collected on May 1st. ................................................... 134
Figure 4-8. Unique and overlapping tweets in the Location, Keyword, and Network
Datasets. ........................................................................................................................... 137
Figure 4-9. Incidents identified by situational reports in the Location, Keyword, and
Network Datasets: power line damage (teal), property damage (yellow), road damage
(pink), storm (orange), internet outage (red), and power outage (blue). .......................... 138
Figure 5-1. Four functions of sensemaking cycle (Gary Klein, Moon, & Hoffman, 2006;
Gary Klein et al., 2007). ................................................................................................... 166
Figure 5-2. Analytics dashboard interface (left) and scene from simulations with
telecommunicators and researchers at mock CAD workstations (right). ......................... 170
viii
LIST OF TABLES
Table 1-1. Research questions for Phases I, II, and III. ........................................................... 21
Table 3-1. Types of information posted on social media during crises. .................................. 39
Table 3-2. Phase I research questions. ..................................................................................... 51
Table 3-3. Interviewed public safety officials by county population. ...................................... 60
Table 4-1. Phase II research questions. .................................................................................... 96
Table 4-2. Categories of community organizations (*unique followers)................................. 115
Table 4-3. Coding categories for infrastructure damage and service disruption. .................... 124
Table 4-4. Self-identified location for users following community organizations. ................. 128
Table 4-5. Situational reports collected in the Location, Keyword, and Network Datasets. ... 135
Table 5-1. Phase III research questions ................................................................................... 160
Table 5-2. Four phases of scenario one.................................................................................... 179
Table 5-3. Three phases of scenario two.................................................................................. 184
Table 5-3. Design requirements supporting sensemaking by social media analysts. .............. 195
Table 6-1. Outstanding questions for incorporating social media in emergency dispatch. ..... 206
ix
ACKNOWLEDGEMENTS
I reserve special thanks for my family to whom I simply express my love. I would like to
express my appreciation and gratitude to my advisor, Dr. Fred Fonseca, for taking me up as a
Masters’ student and patiently working with me since. I also offer my special thanks to my mentors
and colleagues at Penn State and the 3C Informatics Lab, especially Dr. Andrea Tapia, Col. Jake
Graham, Dr. Jess Kropczynski, Shane Halse, Scott Pezanowski, and Bill Aurite. Furthermore, I
would like to thank Dr. Frédérick Bénaben, Dr. Aurélie Montarnal, Dr. Caroline Rizza, and the
faculty, staff, and PhD students at the École des Mines d'Albi. I wish to express my appreciation to
Dr. Jack Carroll and Dr. Clio Andris for serving on my dissertation committee. I also want to thank
Jim Lake and the staff of the Charleston County Consolidated 911 Dispatch Center, as well as Dave
Sehnert, Mike Beagles, and their colleagues at Mission Critical Partners, for organizing and
participating in the design workshops which inform much of this dissertation.
1
Chapter 1
Introduction
Visions of the smart city imagine civic officials managing public services using
sophisticated analytics to collect and process data from an array of physical and social sensors.
Such imaginings include emergency responders intelligently aggregating traditional 911 calls with
“calls” from citizens sending text messages, images, streaming video, as well as social media posts
gathered during a crisis. Within this vision, 911 dispatch centers, known as Public-Safety
Answering Points (PSAPs), transform from reactive call centers to proactive data analytics and
coordination hubs providing first responders with real-time information during, and potentially
before, emergencies.
Reimagining community emergency response finds motivation in the fact that people no
longer only call 911 during an emergency. People increasingly send texts, pictures, and even videos
directly to emergency dispatchers in PSAPs, and, frequently, live-post information on social media
during emergencies that inform local and global networks of social media users. Moreover, people
no longer only wait for emergency alerts broadcast on television, radio, or mobile devices during
an emergency. People search social media to access unofficial information that is often more
immediate, more local, and more abundant than anything provided by official sources. Lastly,
people no longer only wait on emergency response officials for help (and never have): remote,
digital volunteers using social media platforms can help coordinate emergent responses to crises
by gathering and relaying information between locals seeking help and ad hoc citizen responders.
Consequently, for many emergency response officials, only answering 911 calls no longer seems
enough.
2
With the growth of Facebook and Twitter beginning in the mid-oughts, citizens using social
media during an emergency bypass communication channels traditionally connecting citizens and
emergency response officials. On one hand, citizens’ using social media during an emergency
decentralize the emergency reporting channels (e.g. 911) from which officials expect to receive
citizen-reported information during an emergency. On the other hand, citizens accessing
information on social media decentralize the public communication channels (e.g. Emergency Alert
System (EAS) and Wireless Emergency Alerts (WEA)) through which officials share information
with citizens in effort to coordinate community responses to crises.
As a result, the model of centralized emergency response in which government agencies
rely on broadcast media and communications hubs to coordinate the flow of information between
citizens and emergency responders now co-exists with an alternative model of decentralized
information flows and non-professional, citizen responders that emerges during a crisis. Faced with
these competing models, emergency responders wrangle with the problem of coordinating the
inherently informal, decentralized, and recalcitrant behaviors of social media users with the highly-
formal, centralized, and insular processes that have traditionally connected citizens requesting help
with emergency responders capable of providing assistance. For community emergency responders
serving in county and municipal-level jurisdictions, the problem centers on how to incorporate two
new tasks, social media distribution and monitoring, within the activities of emergency responders
organized through the Incident Command System (ICS), the command and control framework that
delegates roles and responsibilities among government agencies and officials to coordinate
emergency response processes at the local, regional, and national levels in the United States.
Significantly, tasks of social media distribution and monitoring resemble and extend
traditional emergency response functions delineated in the ICS. Social media distribution involves
posting and redistributing information on social media platforms in efforts consistent with the ICS
3
public information function: “processes, procedures, and systems for communicating timely,
accurate, and accessible information on an incident’s cause, size, and current situation; resources
committed; and other matters of general interest to the public, responders, and additional
stakeholders” (FEMA, 2013). Conversely, social media monitoring involves collecting and
analyzing social media data to support situational awareness and decision-making among
emergency responders. Within the ICS, social media monitoring resembles the work of Public
Information Officers (PIOs) “reviewing media reports for accuracy, content, and possible response”
(FEMA, 2007, p. 13) and emergency dispatchers answering 911 calls and dispatching first
responders to emergencies (Martini, 2018; Nimsgern, 2018).
Given the analogous functions of public information and emergency dispatch formally
specified and practically carried out among PIOs and emergency dispatchers, the incorporation of
social media in emergency response processes would seem, given some necessary adjustments,
straightforward. Indeed, research in Crisis Informatics has shown that citizens’ not only use social
media to post information during an emergency (Olteanu, Vieweg, & Castillo, 2015; Starbird,
Palen, Hughes, & Vieweg, 2010), but expect officials to monitor and disseminate social media as
well (Petersen, Fallou, Reilly, & Serafinelli, 2018; Reuter & Spielhofer, 2017). Similarly, research
finds that emergency response officials adopt social media during crises (Hughes, & Palen, 2009;
Tapia & Moore, 2014; Y. Zhang, Drake, Li, Zobel, & Cowell, 2015), and, generally, seek to employ
social media to distribute information to the public and, albeit less often, monitor information
reported by citizens (Plotnick, Hiltz, Kushma, & Tapia, 2015; Reuter, Ludwig, Friberg, & Pratzler-
Wanczura, 2015; Reuter, Ludwig, Kaufhold, & Spielhofer, 2016).
However, multiple barriers impede uses of social media among emergency responders
serving in geographic communities, county and municipal jurisdictions within which an ensemble
of officials— first responders in fire, medical, and police departments, emergency managers, and
4
various officials in civic agencies (e.g. public works)— are responsible for cooperatively
preventing, responding to, and managing the impacts of emergencies. Despite general perceptions
of utility and uneven use among emergency response officials serving in various agencies and
jurisdictions (Mergel & Bretschneider, 2013), studies commonly observe a lack of staff with the
time, expertise and training, and requisite policies to employ social media in addition to their
existing duties; a lack of software tools to gather, process, and visualize social media data such that
“actionable” information can be identified without preventing information overload; and, most
commonly, a lack of trust in the accuracy and quality of citizen-reported information on social
media to the extent that officials consider social media unsuitable for operational decision-making
(Hiltz, Kushman, & Plotnick, 2014; Hiltz & Plotnick, 2013, 2016; McCormick, 2016; Plotnick &
Hiltz, 2018; Plotnick et al., 2015; Shan, Plotnick, Hiltz, & Yang, 2017; Y. Zhang et al., 2015). As
a result, a significant gap exists between emergency responders’ perceptions of social media and
uses of social media for distributing and monitoring information (Reuter et al., 2016).
Understanding these barriers requires context. Prior studies observe that PIOs or
emergency managers often lack adequate staff, tools, and trust to distribute or monitor information
on social media (Hiltz & Plotnick, 2016; Hughes, 2014; Hughes, & Palen, 2012). While these
barriers impede social media use among PIOs and emergency managers engaged in public
information work, the tasks of social media distribution and monitoring might be alternatively
delegated among emergency dispatchers in PSAPs in order to directly support first responders
engaged in emergency response. Moreover, employing social media distribution and monitoring in
community emergency response would see new interactions between officials and citizens
traditionally linked through centralized communications channels.
This possibility is important since emergency response involves different cooperative and
coordinative activities than public information work. Examining the lack of staff, tools, and trust
5
within the activity context of emergency dispatch, for instance, allows the reinterpretation of
barriers observed to impede social media use for public information work by considering the
particular constraints and opportunities present in the activity context of community emergency
response (Kaptelinin & Nardi, 2012; Norman, 2005). To examine social media distribution and
monitoring in this context requires special attention to the objectives, cooperative work, and
coordination mechanisms enabling intra and inter-organizational information flows among
emergency responders organized within the ICS, and the coordination mechanisms enabling
information flows between officials’ centralized, ICS-organized processes and citizens’
decentralized uses of social media in the community. As previously described, widespread uses of
social media create new relationships and potential mesh points between the planned and emergent
activities of officials and citizens responding to emergencies in their community.
Examining barriers to social media use in the alternative activity contexts of community
emergency response reveals, in turn, new requirements for coordinating social media distribution
and monitoring during an emergency. Whereas prior research contextualizes these functions within
the related context of public information work outlined in the ICS and performed among PIOs and
emergency managers (Hiltz & Plotnick, 2016; Hughes, & Palen, 2012; St Denis, Hughes, & Palen,
2012), incorporating social media within the interrelated set of activities constituting emergency
reporting, dispatch, and response remains unexamined (Grace, Kropczynski, & Tapia, 2018). In
this activity context, PSAPs play a pivotal role as the central coordination hubs which process
citizens’ 911 calls and disseminate actionable information to emergency responders across local
government agencies in a community (Kropczynski et al., 2018). In this context, the purposes
driving social media distribution and monitoring shift from informing the public to detecting and
aggregating information about emergencies that can support the situational awareness needs of
emergency responders.
6
Importantly, understanding social media use in the context of community emergency
response reframes the barriers impeding social media use in light of broader transformations in the
ways communities report and respond to emergencies. First, when an emergency occurs, multiple
physical and social sensors now provide information to emergency dispatch officials and, in the
case of social media, networks of local and remote citizens (Ajao, Hong, & Liu, 2015; Cuff,
Hansen, & Kang, 2008; Leysia Palen et al., 2010). Whereas citizens have traditionally reported
emergencies by calling emergency telephone numbers such as 911, the adoption of the Emergency
Services IP Network (ESInet) will enable citizens to also send texts messages, images, and video
to PSAPs. In some cases, web-based applications are already available when 911 channels are down
(Holland, 2018). In addition, PSAPs now collect data from an array of physical sensors (e.g.
cameras, license plate readers, gunshot detectors) and search digital records (e.g. arrests,
complaints, summonses, firearm registries) to assist on-scene first responders (Levine & Tisch,
2014). Aligned with visions of the smart city (Cuff et al., 2008; Liu et al., 2015), PSAP officials
now imagine workflows among 911 call takers, dispatchers, and prospective data analysts
responsible for gathering, analyzing, and synthesizing information gathered from physical and
“social sensors” including 911 callers and, potentially, social media users. Such visions imagine
PSAPs transforming from reactive call centers to proactive data analytics and coordination hubs
that can provide first responders with real-time information during, and potentially before,
emergencies (Grace, Kropczynski, et al., 2019).
In this context, incorporating social media requires the integration of information gathered
from social media, 911 callers, and other information sources sustaining processes of reporting and
dispatch organized through the ICS. Uses of social media can transform the direct, one-to-one
communication between a 911 caller and dispatcher into indirect, many-to-one communications
between social media users and social media analysts who, in turn, analyze, aggregate, and
7
synthesize information reported by social media users, 911 callers, and on-scene responders.
Incorporating social media within PSAPs contributes to the transformation of emergency dispatch
around multiple physical and social sensors, the heterogeneous data they create, and future
workflows among emergency dispatch personnel with new specializations.
Second, uses of social media that decentralize emergency communications between
citizens and officials can, in turn, also decentralize emergency response in a community. By
decentralizing the sources and content of information available during an emergency, social media
platforms can disrupt officials’ attempts to coordinate public action by moderating the
dissemination of public information available to citizens during a crisis. In contrast, social media
allows citizens to share and access local information that is often unavailable from official sources
(Dailey & Starbird, 2017; Kogan, Palen, & Anderson, 2015). Thus, while citizens have always
helped each other in times of crisis (Drabek, 1986; Drabek & McEntire, 2003; Quarantelli, 1984;
Stallings & Quarantelli, 1985), social media provides new ways for citizens to coordinate ad hoc
responses at scale. Remote, digital volunteers can curate information resources and coordinate
information flows among affected citizens and on-the-ground professional and non-professional
responders (Ludwig, Kotthaus, Reuter, Dongen, & Pipek, 2017; Leysia Palen et al., 2010; Soden
& Palen, 2016; Starbird & Palen, 2011). Examples such as the “Cajun Navy”, boat-owners who
rescued fellow citizens stranded (and tweeting) on the rooftops of homes flooded during Hurricane
Harvey (CajunNavy, 2017), illustrate appropriations of social media platforms to coordinate an
emergent, citizen response (Kaufhold & Reuter, 2016), especially when traditional emergency
reporting channels such as 911 break down (Hughes et al., 2014).
Consequently, incorporating social media in community emergency response calls for
additional coordination mechanisms compared to traditional public information work. Given the
limited reach of government agency’s social media networks, officials require citizens to re-
8
distribute official messages to the public-at-large (Karanasios, Cooper, Balcell, & Hayes, 2019).
However, in contrast to broadcast channels such as the EAS (i.e. television and radio) and WEA
(i.e. mobile devices), officials employing social media platforms to distribute information to the
public typically do not know who/what is re-distributing information from officials’ to citizens’
social media accounts or, moreover, the citizens in the community who stand to access this
information (Dailey & Starbird, 2017; Grace et al., 2018). In this context, distributing social media
requires awareness of citizens accessing and re-distributing official information, as well as, in the
context of emergency response, awareness of information citizens post during emergencies.
Dissertation Overview
By decentralizing emergency communications and contributing to broader shifts in how
communities report and respond to emergencies, citizens’ social media use encourages emergency
responders to incorporate social media distribution and monitoring within the processes of
community emergency response. With this motivation, the research underpinning this dissertation
proceeds across three stages, moving from the context of community emergency response work to
challenges faced by emergency responders when attempting to gather and disseminate information
on social media, and, finally, to projective opportunities for integrating information from social
media and 911 calls in reconfigured emergency dispatch workflows. Collectively, the three phases
explore opportunities to incorporate social media in ways which enhance community emergency
response.
Phase I examines the incorporation of social media in the context of community emergency
response. By reframing the multiple barriers observed to impede social media use among
emergency responders— a lack of time, tools, and trust— within the activity context of community
9
emergency response points to new opportunities and requirements for coordinating the distribution
and monitoring of social media during an emergency. Across 30 scenario-based interviews,
emergency responders articulate the objectives, cooperative and coordinative work of social media
distribution and monitoring with respect to existing intra and inter-organizational sociotechnical
infrastructures and protocols organizing community emergency response. These interviews suggest
that the lack of staff, tools, and trust impeding social media use among emergency responders can
be overcome by coordinating social media monitoring within emergency dispatch centers, Public
Safety-Answering Points (PSAPs) with unique infrastructures for verifying and integrating citizen-
reported information, and building relationships with “community influencers,” citizens well-
positioned in community networks to redistribute officials’ messages on social media.
Phase II addresses the challenge of awareness: how can emergency responders identify
hyperlocal social media users and collect hyperlocal social media data when distributing and
monitoring information, respectively? To detect reports of emergency in a geographic area requires
new methods for collecting hyperlocal social media data. In the case of Twitter, emergency
responders are limited to the 1-3% of tweets with geographic metadata (i.e. geotags) and tweets
containing keywords (e.g. “flood”) selected to filter the global Twitter stream. While location
filtering excludes the majority of tweets posted to Twitter, keyword filtering excludes any tweet
lacking the narrow set of geographic (e.g. city names) or emergency-related keywords (e.g. flood)
that can make information visible during a global Twitter search. To supplement these methods,
this dissertation introduces Social Triangulation and network filtering to identify hyperlocal social
media users and collect hyperlocal social media data, respectively. In a study comparing social
media data collected using location, keyword, and network filtering methods during a severe
weather emergency, network filtering enables emergency responders to identify twice as many
10
unique reports of infrastructure damage and service disruption as location and keyword filtering
combined.
Lastly, Phase III explores the integration of information from social media users and 911
callers in emergency dispatch work involving 911 call takers, dispatchers, and social media analysts
engaged in processes of distributing sensemaking. As PSAPs already coordinate the flow of
information between citizens and first responders, incorporating social media monitoring requires
911 call takers, dispatchers, and prospective social media analysts to make sense of information
gathered from social media users using information gathered from 911 callers, and vice versa. In
this regard, incorporating social media analytics in emergency dispatch continues the transition of
PSAPs from reactive call centers to proactive data analytics and coordination hubs using Next-
Generation 911 infrastructure to collect and process heterogenous data from physical and social
sensors to support the situational awareness needs of first responders. Phase III describes role plays
and simulations conducted with 911 dispatchers to understand the sociotechnical requirements for
including social media analysts in emergency dispatch workflows. The role plays identify resources
(e.g. SOPs, common information space) that coordinate emergency dispatch work among multiple
actors, technologies, and data sources, and inform the construction of simulations involving a
Computer-Aided Dispatch environment, social media analytics dashboard, and social media and
911 caller datasets for two emergency scenarios. Examining the simulated workflows among 911
call takers, dispatchers, and social media analysts from the perspective of distributed sensemaking,
findings from the simulations point to sociotechnical requirements enabling and coordinating
sensemaking functions— framing, elaborating, questioning, and reframing— performed by
distributed people and systems.
The transitions between Phases I, II, and III reflect the emergent and exploratory character
of this dissertation. Each phase informs the next: interviews with emergency responders during
11
Phase I reveal difficulties accessing hyperlocal social media data, reaching hyperlocal social media
users, and integrating information from multiple social data sources in emergency dispatch centers.
These issues, in turn, motivate Phases II and III. Furthermore, storm-related social media data
collected when comparing location, keyword, and network filtering methods in Phase II create, in
part, the synthetic datasets analyzed by social media analysts during the emergency dispatch
simulations conducted during Phase III. The emergent character of the studies comprising this
dissertation are considered in the conclusions which presents the use of low and high-fidelity design
enactments in this dissertation as a methodological approach for the design of and for activities in
emergency response and management. Furthermore, this dissertation concludes that the activity
context of community emergency response, characterized by distributed sensemaking, information
integration, collaborative analytics, and multi-sensor systems, require a shift in Crisis Informatics
research from studies examining uses of social media per se to examinations of the sociotechnical
processes by which emergency responders and citizens make sense of heterogenous information
within increasingly data-rich emergency environments. Phases I, II, and III are described in further
detail below.
Phase I: Context
At the same time as citizens turn to social media to access and share information during
emergencies, persistent barriers impede emergency responders from using social media to
distribute timely information to the public and monitor information that can support the situational
awareness needs of officials and citizens. Emergency responders serving in communities-
emergency managers (Hiltz et al., 2014; Hiltz & Plotnick, 2013; Plotnick et al., 2015; Shan et al.,
2017), Public Information Officers (PIOs) (Hughes, 2014; Hughes & Palen, 2012; Hughes & Shah,
12
2016), emergency dispatchers (Boersma, Diks, Ferguson, & Wolbers, 2016), and first responders
(fire, medical, police) (Kaufhold & Reuter, 2017; Kavanaugh et al., 2012; Reuter et al., 2016)-
commonly report a lack of staff and available time to make use of social media, a lack of appropriate
tools and expertise to collect and process voluminous social media data, and a lack of trust in the
accuracy and quality of citizen-reported information which renders social media unsuitable for use
in operational decision-making (Hiltz & Plotnick, 2016). However, whereas prior work observes
these barriers among individual officials and individual agencies, the community-level activity
context(s) for incorporating tasks of social media distribution and monitoring, the set of cooperative
and coordinated tasks characterizing emergency response processes, remain undefined.
Contextualizing uses of social media in the activity context of community emergency response can
reveal coordination mechanisms that suggest (re-)delegating and resourcing these tasks in
cooperative arrangements that might overcome observed barriers to social media use (Grace et al.,
2018).
Community emergency response consists of the cooperative and coordinative activities
performed by professional and non-professional actors and organizations in a geographic
community immediately before, during, and after an emergency to mitigate harmful social and
environmental impacts. This context is important as prior research examines social media use as
public information work (Hughes, & Palen, 2012; St Denis et al., 2012), and focuses on individual
and organizational-level barriers—a lack of staff, tools, and trust— among emergency responders,
including emergency managers (Hiltz et al., 2014; Hiltz & Plotnick, 2013; Plotnick et al., 2015;
Shan et al., 2017), Public Information Officers (PIOs) (Hughes, 2014; Hughes & Palen, 2012;
Hughes & Shah, 2016), emergency dispatchers (Boersma et al., 2016), and first responders (fire,
medical, police) (Kaufhold & Reuter, 2017; Kavanaugh et al., 2012; Reuter et al., 2016).
13
Consequently, these studies often ignore cooperative and inter-organizational uses of social media
that characterize community emergency response work (Grace et al., 2018).
Accordingly, Phase I conducts scenario-based interviews to examine the objectives and
cooperative and coordinative work practices of emergency responders using social media to
distribute and monitor information in the activity context of community emergency response.
Interviews were conducted around use-case scenarios, “short stories about hypothetical characters
in specified circumstances, to whose situation the interviewee is invited to respond” (Finch, 1987).
The scenarios prompted emergency responders to consider the distribution and monitoring of social
media in cooperation with citizens, community volunteers (to compensate for a lack of staff), and
officials in different community agencies in concrete situations of use (Carroll, 2000, 2015).
Examining emergency responders’ reflections on these scenarios through the perspective of
cooperative and coordinative (i.e. articulation) work, enabled the examination of social media use
as a set of mutually-dependent activities coordinated among community actors and agencies
employing local, available resources (Schmidt, 2008; Schmidt & Bannon, 1992; Schmidt &
Simonee, 1996).
Across these interviews, emergency responders discuss cooperative work arrangements for
social media distribution and monitoring in which citizens, officials, and community volunteers
perform clearly-defined roles. First, and principally, officials use social media as a mass broadcast
system for distributing timely messages to the public. By distributing official information via social
media, emergency responders seek to coordinate public action during an emergency. Social media
distribution, however, requires aligning the activities of official “gatekeepers” posting information
with citizens seeking information in emergency situations. To support this coordination, emergency
responders require the help of “community influencers,” citizens well-positioned in community
networks to disseminate officials’ messages on social media. While emergency responders seek to
14
reach both broader and targeted demographics with their messages, few officials know if local
citizens are engaging information posted on social media or attempt to involve citizens-as-
community influencers in emergency communication planning.
Second, and discussed primarily among emergency responders in urban, populous
jurisdictions, social media can function as an indirect and distributed reporting system providing
information that can supplement direct reports gathered from 911 callers. In this respect, social
media offers emergency responders with early warning of events and additional situational
awareness information during an emergency. For these officials, social media monitoring allows
dispatchers to proactively detect emergencies, more quickly dispatch first responders, and provide
them with more information than when relying on 911 callers alone. Social media monitoring,
however, requires incorporating social media analytics in PSAPs, community agencies uniquely
positioned to verify and synthesize information posted on social media for dispatch to first
responders.
This study provides new perspectives on observed barriers to social media use among
emergency responders. Contrasting research focusing on emergency managers and PIOs as the
primary adopters of social media analytics (Hiltz et al., 2014; Hiltz & Plotnick, 2016; Shan et al.,
2017), the scenario-based interviews point to emergency dispatchers serving in PSAPs as critical
adopters. However, PSAPs require social media analytics software and procedures that can
facilitate the collection, discovery, and verification of actionable information on social media and
support the aggregation and synthesis of information from social media users and 911 callers.
Furthermore, observing the need to develop ties between emergency responders and citizens well-
positioned to redistribute official messages on social media suggests an analog and remediation of
traditional PIO relationships with local television and radio stations (Hughes & Palen, 2012).
Looking to “community influencers” as additional emergency communication channels available
15
to emergency responders presents new opportunities for citizens to participate in public safety that
can bolster the social capital of communities during emergencies (FEMA, 2011; Grace et al., 2019;
Murphy, 2007).
Phase II: Awareness
As citizens often use social media platforms such as Twitter to access and share critical
information during emergencies (Mazer et al., 2015; Olteanu et al., 2015; Starbird et al., 2010),
emergency responders use social media to distribute timely information to citizens and detect
information contributing to situational awareness (Denef, Bayerl, & Kaptein, 2013; Grace et al.,
2018; Hiltz & Plotnick, 2016; Hughes, & Palen, 2012). However, existing methods to collect real-
time social media data during a crisis remain limited to location and keyword filtering despite the
sparsity of geographic metadata associated with tweets, and the tendency of keyword-based
methods to capture highly-visible information posted by remote rather than local users (Bruns &
Liang, 2012; Carley, Malik, Landwehr, Pfeffer, & Kowalchuck, 2016; Morstatter, Pfeffer, Liu, &
Carley, 2013; Olteanu, Castillo, Diaz, & Vieweg, 2014). As a result, community emergency
responders end up collecting only a small proportion of hyperlocal data created by a small
proportion of citizens in a geographic area.
The limitations of location and keyword filtering demand new methods for identifying
hyperlocal social media users and collecting the data they create in order to effectively distribute
official information and identify situational awareness information during an emergency. This
study introduces Social Triangulation and network filtering as novel methods for inferring
community networks of social media users located in a geographic area and collecting hyperlocal
social media data created by community networks, respectively. Phase II first describes and
16
evaluates the method of Social Triangulation by inferring networks of Twitter users in State
College, Pennsylvania (Grace et al., 2017), and, second, compares data collected using location,
keyword, and network filtering (via Social Triangulation) methods during a severe weather event,
(including an F1 tornado), that took place in State College on May 1st, 2017 (Grace, Halse, Aurite,
Montarnal, & Tapia, 2019).
First, Social Triangulation is introduced as a method to infer local users vis-à-vis the local
organizations they follow on Twitter. Based on the hypothesis that local citizens are more likely to
follow the Twitter accounts of local organizations in their communities than non-local citizens,
Social Triangulation involves cataloguing and categorizing the accounts of local organizations to
infer community networks among Twitter users who tend to follow multiple, local organizational
accounts. The method is then evaluated using users’ self-identified profile locations. Among 79,978
Twitter users following a local organization in State College and including location information in
their profiles, 54,165 (68%) self-identify as living in State College. Moreover, the more local
organizations users follow the more likely they are to self-identify as local citizens. For users
following one or two local organizations, 37,788 or 67% identify as local citizens. This increases
among those who follow 3-9 organizations (71%), 10-49 organizations (84%), and, among those
following 50 or more organizations, 98% self-identify as local. In addition, by categorizing the
Twitter accounts of local organizations, Social Triangulation identifies important sources and types
of community information followed among local Twitter users. In the case of State College, local
media accounts reach the most users, however, distinct “filter bubbles” exist among networks of
users following a single category of organizational account(s) (e.g. schools, civic services, bars,
etc.).
Second, network filtering is introduced to complement existing location and keyword
filtering data collection methods. Network Filtering collects social media data from networks of
17
Twitter users associated with a geographic community (via Social Triangulation). Furthermore, in
the case of a severe weather emergency, data collected using location, keyword, and network
filtering are compared by analyzing the distribution of situational reports of infrastructure damage
and service disruption across location, keyword, and network-filtered social media data collected
during a storm which saw flooding and the touchdown of an F1 tornado in Centre County,
Pennsylvania. Findings reveal that location and keyword filtering respectively identify 28% (n=97)
and 20% (n=72) of all situational reports collected across the three methods (n=352). In contrast,
network filtering doubles, 52% (n=183), the number of situational reports collected in real-time
compared to location and keyword filtering alone, suggesting that typical methods for collecting
social media data during a crisis capture only a fraction of all social media data created by citizens
in affected communities.
The complementary methods of Social Triangulation and network filtering offer two
primary contributions for collecting social media data and improving situational awareness during
an emergency. First, by inferring networks of local social media users and collecting data they
create, network filtering increases the volume of hyperlocal social media data available to
emergency responders and, in turn, can expand awareness of incidents occurring in a community.
However, while network filtering identifies nearly three quarters (73%) of all incidents reported
during the emergency, each of the three methods identify unique incidents of infrastructure damage
and service disruption reported on Twitter. These findings suggest that combining location,
keyword, and network filtering methods provide a complementary suite of data collection methods
to support situational awareness during a crisis. Second, by describing networks of social media
users associated with a geographic community, Social Triangulation can inform emergency
communications planning by allowing emergency responders to identify “community influencers”
positioned to effectively redistribute official messages to the widest audience possible. In this
18
regard, this study offers emergent guidelines for employing social media in emergency
communications planning: account for local “filter bubbles,” identify community influencers, and
cooperate with citizens’ associations.
Phase III: Integration
While research suggests that social media can enhance the situational awareness of
emergency responders during a crisis (Q. Huang & Xiao, 2015; Rudra et al., 2016; Saleem, Xu, &
Ruths, 2014; Starbird et al., 2010; Vieweg, Hughes, Starbird, & Palen, 2010; Zade et al., 2018),
few studies address the process of distributed sensemaking in which social media data created by
citizens are transformed into actionable information for first responders. Through scenario-based
role plays and simulations conducted with 911 dispatchers serving in a Public-Safety Answering
Point (PSAP), this study examines the distributed sensemaking processes that emerge when
incorporating social media analysts and analytics within the emergency dispatch workflows.
Workshops were held in May and August 2018 in which 911 dispatchers took part in role
plays imagining the inclusion of social media analysts in emergency dispatch work and simulations
involving 911 call takers, dispatchers, and prospective social media analysts using Computer-Aided
Dispatch (CAD) systems and a social media analytics dashboard during mock emergency scenarios.
During the first workshop, dispatchers participated in role plays and created, through improvisation,
both the emergency scenario and information reported by 911 callers and social media users.
Widely used in human-centered design, role playing allows researchers to observe plausible
interactions among domain-experts and projective end-users as they develop within hypothetical
situations (Medler & Magerko, 2010; Simsarian, 2003; Valkonen & Liinasuo, 2010). Here role play
allowed dispatchers to imagine cooperating with a social media analyst and, at the same time,
19
plausibly ground their projective interactions within existing work practices and procedures (Grace,
Kropczynski, et al., 2019). The results of the first workshop informed the construction of
emergency scenarios and synthetic 911 call and social media datasets that formed the basis for
emergency dispatch simulations conducted during the second workshop.
Findings from the role plays reveal that information gathered from social media might
address information gaps that emerge when 911 callers fail to provide critical information and vice
versa, suggesting social media enhances situational awareness only when integrated into
sensemaking processes that synthesize information across multiple, incomplete, but
complementary data sources. This synthesis, however, requires cooperative information gathering
and sharing among call takers, dispatchers, and social media analysts that PSAPs can coordinate
using common interpretive frameworks (e.g. 6W protocol) and common information spaces (e.g.
CAD). As a result, information gathered from social media data can enhance situational awareness
only by aligning information outputs of social media with coordination mechanisms organizing
distributed sensemaking processes.
Extending these findings, the simulations demonstrate the distribution of sensemaking
tasks among call takers, dispatchers, and analysts, and how distributed sensemaking processes can
be organized around the distinct domain-ontology organizing emergency dispatch work. This
distributed process begins when the call taker or analyst “frames” the situation by using CAD as a
common information space to share the primary attributes of the emergency (i.e. incident type and
location). This frame then provides a set of information requirements that coordinate information
seeking and integration among the call taker, dispatcher, and analyst who attempt to discover,
aggregate, and synthesize situational awareness information required by first responders. When
inconsistencies arise among information from social media and 911 callers, dispatchers compare
among and between available primary and secondary information to elaborate the existing
20
emergency frame or reframe the situation by entering an additional call in CAD. The simulations
reveal, however, that social media analysts will likely require event detection systems and alerts to
prompt information seeking, assisted search capabilities to increase the recall and precision of
analysts’ search queries, and new protocols that coordinate priority information seeking to address
information gaps that evolve during an emergency response.
Overall, the role plays and simulations reveal how the distinct domain ontology organizing
emergency dispatch work coordinates sensemaking among multiple actors engaging multiple
technologies and data sources. Findings contribute to theory surrounding social media,
sensemaking, and situational awareness by showing that social media content does not, ipso facto,
enhance situational awareness in emergency response unless coordinated within the distributed
sensemaking processes of emergency responders (Baber & McMaster, 2016; Kropczynski et al.,
2018; Zade et al., 2018). Social media cannot be simply “pumped in” to officials but must be
coordinated within existing workflows in which it provides incomplete information only in relation
to other incomplete information sources. Situational awareness is, therefore, the achievement of
domain-dependent processes that coordinate the integration of information across multiple,
incomplete, but complementary data sources to meet unfolding information requirements during
an emergency. To meet these requirements, the simulations suggest that social media analysts will
require new coordinative protocols and social media analytics software to proactively address
evolving information needs during an emergency.
Outline of Chapters
Chapter One describes how the decentralization of emergency communications between
community emergency responders and citizens motivates the three phases of research that
21
constitute this dissertation. Chapter One outlines the research question orienting each phase of
research and summarizes their primary findings and contributions.
Chapter Two, “Community Emergency Response,” defines the context of this research: the set of
cooperative and coordinated activities performed by professional and non-professional actors in a
geographic community immediately before, during, and directly after an emergency to suppress its
cause and mitigate its human and environmental impacts. Chapter Two defines the concepts of
“community,” “emergency,” and “response,” and provides background information on the Incident
Command System, the 911 system, and Public-Safety Answering Points.
Chapter Three, “Context,” reframes the barriers- lack of staff, tools, and trust- impeding social
media use among community emergency responders within the activity context of emergency
dispatch to highlight the coordinative roles of community influencers and PSAPs in social media
distribution and monitoring, respectively. Presenting findings for Phase I, Chapter Three outlines
challenges of awareness and integration addressed in Phases II and III (Table 1-1).
Table 1-1. Research questions for Phases I, II, and III.
Phase Research Question
I What objectives, cooperative tasks, and coordinative practices characterize social media
use in community emergency response?
II How can emergency responders identify hyperlocal social media users and collect
hyperlocal social media data?
III How can emergency responders integrate information from social media and 911 callers
in emergency dispatch operations?
22
Chapter Four, “Awareness,” introduces the dual methods of social triangulation and network
filtering to infer hyperlocal community networks and use these networks to collect hyperlocal social
media data, respectively. Presenting findings for Phase II, Chapter Four demonstrates how social
triangulation and network filtering enable emergency responders to identify community influencers
that can reach “filter bubbles” among social media users in a community and collect data that
expands opportunities for situational awareness during an emergency.
Chapter Five, “Integration,” describes role plays and emergency dispatch simulations conducted
with 911 telecommunicators to understand the sociotechnical requirements for including
prospective social media analysts in emergency dispatch workflows. Presenting findings for Phase
III, Chapter Five describes existing and required sociotechnical resources that can enable
distributed sensemaking in next-generation emergency dispatch work.
Chapter Six summarizes the contributions of Phases I, II, and III and describes directions for future
work. This dissertation advances Crisis Informatics research from questions concerning how social
media may contribute to situational awareness to the domain-specific processes of sensemaking
that condition the “actionability” of social media in community emergency response. Furthermore,
by employing, first, low-fidelity enactments to understand future emergency dispatch workflows
and, second, high-fidelity enactments to understand sociotechnical design requirements enabling
these workflows, this dissertation demonstrates a methodological approach for co-designing future
activities and artifacts with crisis responders.
23
Lastly, the back matter lists references and includes appendices for the semi-structed interview
protocol employed during Phase I, Institutional Review Board approval for the studies conducted
in this dissertation, and funding information.
24
Chapter 2
Community Emergency Response
Community emergency response consists of cooperative and coordinated activities
performed by professional and non-professional actors in a geographic community immediately
before, during, and after an emergency to mitigate and suppress human and environmental
impacts. Distinct from disasters, emergencies occur within, impact, and are resolved by people
living in a geographic community. Distinct from emergency management functions of preparation,
mitigation, and recovery, emergency response consists of “activities taken immediately before,
during, or directly after an emergency that save lives, minimize property damage, or improve
recovery” (McLoughlin, 1985, p. 166). Lastly, distinct from international, national, and regional
government and non-government crisis and humanitarian responders, community emergency
responders include police officers, fire fighters, emergency medical technicians, emergency
managers, 911 telecommunicators (including call takers and dispatchers), Public Information
Officers (PIOs), and various civic officials serving in geographic communities coextensive with
county and municipal-level jurisdictions. While Phase I concerns the delegation of cooperative and
coordinative tasks of social media distribution and monitoring among this ensemble of emergency
responders, Phase III focuses on the work of emergency dispatchers as the critical coordination
work necessary for translating social media data into actionable, situational awareness information
useful for the full ensemble of emergency responders serving a community.
25
Distinguishing Emergencies, Disasters, and Catastrophes
The scale of a crisis defines the purview of emergency response and management work.
First, what is an emergency? Curiously, “emergency management” traditionally does not concern
emergencies. As Rotanz (2007) explains:
An emergency is a relatively routine incident that does not have community-wide impact,
does not require the extraordinary use of resources or procedures to bring conditions back
to normal, can be successfully handled by local emergency responders, is manageable and
clearly defined, is brought under control quickly, and generally involves responders who
know one another. In addition, roles and responsibilities in an emergency are clear-cut, and
there is a recognizable authority structure. (p. 144)
As explained, emergencies remain the purview of emergency response personnel- medial, fire, and
police professionals. A car accident, for instance, most often constitutes an emergency and as such
would not enter the professional jurisdiction of emergency management. However, if the accident
is of sufficient size as to produce consequences beyond those affecting the people involved (i.e. car
passengers), and beyond the capacities of emergency responders to manage, local emergency
managers become involved.
An important distinction will be made between routine emergencies (here referred to as
incidents) and less routine emergencies (referred to here as simply emergencies) producing indirect
consequences for others, and thus defined by community-wide impacts. Here emergency
management work begins:
When a disaster [or, here, emergency] occurs, however, police, fire, and emergency
medical service (EMS) personnel cannot always cope with the resulting widespread
impacts unless an emergency manager and numerous others are available to acquire
resources for first responders and take care of broader response and recovery needs in the
community (e.g., warning, sheltering, debris management, donations management,
rebuilding, etc.). (McEntire, 2007, p. 169)
Returning to the previous example of the traffic accident, if the accident causes traffic interruptions
that stand to significantly interrupt public services (e.g. bus system) or activities in the community
26
(e.g. children returning home on school buses, a community parade, rush hour traffic, etc.) such an
emergency becomes addressed by emergency management officials and organizations.
In contrast to incidents and emergencies, Quarantelli (2000) distinguishes a disaster on four
points. In contrast to the former, disasters see a convergence of non-local response organizations
and the resultant creation of unfamiliar, cooperative relationships, the restriction of local and
individual organizational autonomy as a result of these relationships, a shift to new performance
standards to address compressed time and resource constraints, and, lastly, closer cooperation
between public organizations and private organizations and everyday citizens. As Quarantelli
explains, this last point, regards “the need for the quick mobilization of resources for overall
community crisis purposes” such that private resources often becomes required, and either
requisitioned or volunteered” (p. 1-2).
In turn, a disaster differs from a catastrophe. According to Quarantelli (2000), catastrophes
differ because they involve the loss or destruction of built infrastructure, the inability of local
officials do perform their work due to the loss or destruction of local response capacities, the
simultaneous interruption of most community functions, and the inability of nearby communities
to provide aid as they are similarly debilitated. Catastrophes see the immobilization of both local
and regional response personnel and resources, thus in lieu of the defining emergency management
responsibility during disaster response- coordinating local and non-local response organizations- is
added the now primary challenge of mobilizing and importing remote professional response
personnel and resources.
Focusing on periods of stability concerns neither disaster nor catastrophe (together to be
regarded as crises) but what are here referred to as incidents and emergencies. The functions of
emergency management thus can be differentiated according to the scale of emergency, where local
emergency management during periods of stability involves, at the same time, a primary effort to
27
(1) mitigate community emergencies as well as efforts (2) preparing for, responding to, and
recovering from community emergencies, (3) as well as the charge to mitigate and prepare for the
possibility of crisis, and respond and recover from a crisis should one occur.
Importantly, the functions addressing incidents, emergencies, adisasters and catastrophes
are not discrete but overlap in the local capacities by which communities respond to all three.
Critically, the development of these capacities occurs recursively. That is, the resources by which
communities mitigate and respond to emergencies are (locally) the same as those mitigating and
responding to crisis. Moreover, the way a community organizes personnel and resources in the
mitigation and response of emergencies develops the capacities by which communities address
crises.
From another perspective, emergencies constitute occasions for both cooperative work
among community emergency response officials and citizens that, at the same time, articulates the
cooperative work arrangements for the prevention, response, and recovery from future crises. This
recursive infrastructuring of capacities emergency management that develops through successive
experiences of emergency response and management in lieu of a disaster represents the distinct
feature of local emergency response and management work and the underlying mechanism of
community resilience-building.
Emergency Management Cycle
Emergency management refers to the practice and discipline of emergency management
and response professionals and their institutions, as well as a function of communities generally,
involving both professionals and everyday citizens alike. As a profession, emergency management
consists of officials at every level of government, from small municipality (e.g. boroughs and
28
townships) and county-level emergency managers, to state and national-level executives, who are
responsible for planning and coordinating how emergency response organizations including
medical services (e.g. EMS), fire and police departments, and NGOs (e.g. Red Cross), can most
effectively prevent and respond to emergencies.
Emergency management is typically described to consist of four phases and disciplines
(Haddow, Bullock, & Coppola, 2014; McLoughlin, 1985; Petak, 1985; Waugh & Tierney, 2007):
mitigation, preparation, response, and recovery. Mitigation involves the prevention of hazards by
assessing and addressing risks. “Mitigation attempts to eliminate hazard risk by reducing either the
likelihood or the consequences of the risk associated with the particular hazard” (Haddow et al.,
2014, p. 106). Mitigation entails hazard mitigation planning and risk communication notifying the
public of local risks, as well as engaging the public in identifying, planning, and monitoring risk
management programs. The latter often involves long-term community efforts focused on large-
scale hazards that include land development regulations, building codes, tax subsidies, and public
education programs (Godschalk, 2007, p. 111; McLoughlin, 1985, p. 166).
Preparation occurs when threats cannot be prevented. Preparedness “seeks to improve the
abilities of agencies and individuals to respond to the consequences of a disaster event once the
disaster event has occurred” (Haddow et al., 2014, p. 106). For emergency management
professionals, preparation centers on the development of an emergency operations plan (EOP)
individualized according a community's vulnerabilities and capacities, and standard operating
procedures (SOPs) that guide responders in critical response tasks (Lindell & Perry, 2007, p. 139).
Together, the phases of mitigation and preparation precede the outbreak of an emergency event and
constitute the primary functions for community resilience-building.
Response consists of the “activities taken immediately before, during, or directly after an
emergency that save lives, minimize property damage, or improve recovery” (McLoughlin, 1985,
29
p. 166). Response activities can include “warning, evacuation, search and rescue, emergency
medical care, fire suppression, and other methods to care for disaster victims and minimize
disruptions” (McEntire, 2007, p. 172). Importantly, emergency management officials do not engage
in response activities directly, but “stand up” emergency operations centers (EOCs) to support and
coordinate emergency response services according to pre-existing EOPs. As discussed previously,
emergency response remains the focus for most research, especially crisis informatics studies
addressing social media use in emergency management work.
Recovery refers to a complex process, but can be generally considered as “a period of time
where deliberate actions are undertaken to routinize everyday activities of those individuals and
groups whose daily routines have been disrupted. These activities may restore old patterns and/or
institute new ones” (Quarantelli, 1999, p. 3). Recovery involves restoring and improving the
physical and institutional (e..g public services) infrastructure, as well as the social and
environmental conditions of a community. Recovery spans both short-term efforts that overlaps
with emergency response operations (e.g. temporary housing), and long-term recovery that
involves the fundamental needs of a community: housing, transportation infrastructure, and basic
utility services (e.g. water, electricity, telephone/internet) (Phillips & Neal, 2007, p. 208).
The four phases or functions provide a neat description of the complex and overlapping
sets of emergency management practices. The notion of phases, however, necessarily describes a
linear process that conceals particular issues related to the scale of emergencies, and the multi-
scaled character of emergency management functions that define community emergency response
and management work in particular.
30
Whole Community Approach
Following FEMA’s vision of a “Whole Community” approach, emergency management is
presented as a form of community work, performed between the public and emergency
management and response officials, and coextensive with the activities of community life. Local
emergency management work is distinguished by the simultaneous management of emergencies
while preparing for large-scale crises. The relationship between these two processes becomes the
space of resilience-building, such that the capacities involved in addressing emergencies develop
the capacities by which a community mitigates, prepares for, responds to, and recovers from crises.
As this expands the scope of hazards facing local community emergency management officials, the
management of risks associated with local emergencies becomes central to community emergency
management and resilience-building.
The “Whole Community” approach begins with a proposition: “A community-centric
approach for emergency management that focuses on strengthening and leveraging what works
well in communities on a daily basis offers a more effective path to building societal security and
resilience” (FEMA, 2011, p. 22). The approach recognizes the necessary participation of everyday
citizens in cooperation with officials to enable all four functional phases of of emergency
management and thus build resilient communities. According to FEMA,
Whole Community is a means by which residents, emergency management practitioners,
organizational and community leaders, and government officials can collectively understand and
assess the needs of their respective communities and determine the best ways to organize and
strengthen their assets, capacities, and interests. By doing so, a more effective path to societal
security and resilience is built. (FEMA, 2011, p. 3)
The Whole Community approach presents a vision, a way “to think about conducting
emergency management” (p. 3). This study explores Whole Community emergency management
31
as a form of community work (Schafer, Carroll, Haynes, & Abrams, 2008) by understanding how
citizens and officials can cooperate in communicating local risks in order to enhance community
risk awareness.
The concept of community, however, still need to be defined. Community here simply
refers to people in the same proximate, geographic space. These spaces often find demarcation in
administrative and municipal borders that serve as the jurisdictional boundaries for public services
provided by public institutions. Local emergency management agencies constitute only one such
institution and service provider.
Local emergency response and management usually involves the emergency management
or public safety services division of county and town governments, though organizational structures
and titles vary (McEntire, 2007). Consequently, one way to define emergency response and
management follows from the public mandates of these agencies. “Within the context of the various
statutes, regulations, and ordinances,” Petak (1985) explains, “…emergency management can be
defined as the process of developing and implementing policies that are concerned with…
mitigation... preparedness... response… recovery” (p. 3).
Another perspective is provided by Blanchard et al. (2007) who describe emergency
management as “the managerial function charged with creating the framework within which
communities reduce vulnerability to hazards and cope with disasters.” (Blanchard et al., 2007, p.
4). This definition signals that emergency management involves more than public administrators
in cooperation with officials in police, medical, and fire departments. Instead it recognizes those
many collaborative forms of action by which people in communities safeguard themselves, others,
and the environment- not only during crises- but during everyday periods of stability.
During periods of stability, and reflecting the symmetry underlying resilience, the scope of
emergency management necessarily broadens to recognize how communities do more than prepare
32
for disaster. Haddow, Bullock, and Coppola (2013) define emergency management as “a discipline
that deals with risk and risk avoidance” (p. 2). While emergency management necessarily concerns
large-scale crises including natural and human-made disasters, focus on emergencies (both as
incidents and emergencies per se) and preparation and mitigation (e.g. risk management) extends
the purview of emergency response and management to a broad range of potential events
concerning communities. As such, “emergency management is integral to the security of
everyone’s daily lives and should be integrated into daily decisions and not just called on during
times of disaster” (p. 2).
Community Emergency Response
Community Emergency response involves the efforts of citizens and officials in geographic
communities, typically county and municipal jurisdictions, to prevent and respond to routine
incidents and emergencies, events distinguished by the limited scope of their impacts and response.
While the impacts of incidents and emergencies remain local, affecting people within a geographic
community, and remain the responsibility of community officials and citizens, the scope and
impacts of disasters cannot be resolved by community resources alone: outside assistance becomes
required (Quarantelli, 2000). Thus, the people, natural and built environment of communities
constitutes the unique context of community emergency response.
Consistent with the Whole Community vision, this study approaches community
emergency response with a community framework that necessarily includes emergency response
officials as well as citizens and citizens’ organizations. Community emergency response includes
local citizens who experience, respond to, and report incidents using emergency telephone numbers
monitored by telecommunicators serving in local 911 call centers or Public Safety Answering
33
Points (PSAPs); first responders (police, fire, medical) dispatched to these incidents; emergency
managers responsible for emergency operations planning and risk management surrounding
community events (e.g. festivals, sporting events), and emergency response operations during
natural (e.g. flooding) and man-made (e.g. protests) emergencies; Public Information Officers
(PIO) responsible for communications between municipal agencies and the public; as well as
members of various civic agencies (e.g. public works) and community groups (e.g. red cross) that
provide assistance during an emergency.
Here “community emergency responders” will denote officials serving in local government
agencies. These include officials in governmental agencies within jurisdictions at county or
municipal-level (or lesser) administrative, territorial divisions (boroughs, townships, etc.) (Choi &
Brower, 2006; McGuire & Silvia, 2010; Waugh, 2007). However, “community emergency
response” as a process consists of both professionals and non-professionals (i.e. citizens)
participating in the activities of emergency response. Consequently, community emergency
response consists of the set of cooperative and coordinated activities performed by professional and
non-professional actors in a geographic community immediately before, during, and directly after
an emergency to suppress its cause and mitigate its human and environmental impacts.
Emergency Reporting and Dispatch
Community emergency response revolves around 911, the “universal emergency number”
citizens can call to request emergency services. Calling 911 connects the caller with an emergency
call taker, also referred to as an emergency telecommunicator, serving in a Public-Safety
Answering Point (PSAP). The call taker will question the caller on the nature and location of the
emergency and, either directly or via a separate dispatcher serving in the PSAP, dispatch the
34
appropriate first responders- fire, medical, and/or police- to the scene (NENA, 2018a). As such, the
PSAP is responsible for translating citizen reports of emergencies into actionable, situational
information for emergency responders in a community jurisdiction.
The 911 emergency telephone was first established in the United States in 1968 when
AT&T announced that the company had selected 9-1-1 as the emergency code that would be
implemented throughout the United States. The 911 system was adopted throughout the United
States among state and local jurisdictions, however, in 1976 only 17% of the American population
had 911 service. In 1987, service had extended to 50% of the population and, by the end of the
century, 96% of the population could access 911 (NENA, 2018a).
As cellular services expanded needs arose to enhance 911. In 1996, after a growing number
of incidents prompted officials to address the challenge of locating 911 callers using call phones,
the Federal Communications Commission (FCC) announced an Enhanced 911 (E-911) initiative to
incorporate location service for all cell phones by 2001 (Reed, Krizman, Woerner, & Rappaport,
1998). Again, in 2010, the FCC called for increased accessibility to 911 services by providing SMS
text-to-911 capabilities nationwide. In 2012 the four largest carriers (Verizon, AT&T, Sprint, and
T-Mobile), the National Emergency Number Association, and the Association of Public-Safety
Communications Officials (APCO) agreed to provide an interim, nationwide text-to-911 service by
2014 (NENA, 2018b).
Early outlined in NENA’s “Future Path Plan” in 2001, spurred by the 911 Improvement
Act of 2008, and pilot program led by the Department of Transportation (NHTSA, 2018), the 911
system began changing again with the Next-Generation 911 (NG 911) initiative: the transition of
911 services to an IP-based network (Emergency Services Internet Protocol Network or ESInet)
that will enable citizens to communicate via voice, text, or video, and associated GPS location
information, with a PSAPs which, in addition, will also be able to receive data from various
35
physical sensors in buildings (e.g. security cameras, alarms) vehicles (e.g. Advanced Automatic
Collision Notification) and personal medical devices (911.Gov, 2016). The transition to NG-911
outlines the technical capacities for PSAPs and the potential for networks of citizens, as “social
sensors,” and numerous physical sensors to communicate data during an emergency. However, how
this data will be collected, processed, and visualized, and made sense of among call takers,
dispatchers, and prospective data analysts in PSAPs, as well as the emergency responders with
whom they communicate, remains much less defined.
36
Chapter 3
Phase I: Context
For over a decade, research in Crisis Informatics has examined the use of Information and
Communications Technologies (ICTs), and especially social media, during disasters and
emergencies. This growing body of research, however, has so far favored investigations into the
coordination of social media use among citizens, remote volunteers, and crisis responders during
large-scale disasters. In contrast, much less attention has been directed to the coordination of social
media use in geographic communities beset by intermittent and more-or-less routine emergencies.
This neglect of attention is significant as studies addressing community emergency response
consistently observe barriers- a lack of staff, tools, and trust- impeding social media use among
emergency responders. Consequently, while studies in Crisis Informatics note the need for social
media use in community emergency response, studies have left unexamined the i) the tasks, and
resources necessary for incorporating social media distribution and monitoring in the activity
contexts and processes of emergency response among emergency responders and citizens, and ii)
the coordination mechanisms required for incorporating the tasks and resources for social media
distribution and monitoring within these processes.
Literature Review: Coordinating Social Media Use During Crises
During a crisis, people post, modify, and redistribute information that become resources
for situated action among citizens and crisis responders. Starbird and colleagues (2010) observe
three information behaviors among social media users during a crisis. First, generative information
production occurs when users post first-person observations or status updates of crisis-related
events (Mazer et al., 2015; Saleem et al., 2014; Vieweg et al., 2010). These users post information
37
on social media that becomes a “backchannel” to information broadcast by traditional news media
and government agencies (Shklovski, Palen, & Sutton, 2008). Although constituting only 10% of
all social media posts, Starbird et al. (2010) observe generative information to be disproportionately
posted by users located near and affected by the impacts of the disaster (p. 246). Similarly, across
26 different disasters, Olteanu et al. (Olteanu et al., 2015) observe that eyewitness accounts account
for, on average, 9% of all disaster-related social media posts.
Generative information, in turn, resources the activities of other local and remote social
media users: “generative information is at the core of the information production cycle, providing
the raw material that later production behavior works to shape into a meaningful informational
resource” (Starbird et al., 2010, p. 246). Synthetic information behavior finds users “digesting,
filtering, relaying, and adapting” information discovered on social media or obtained from external
sources, such as news agencies. Users synthesizing information play an important role by
summarizing information from across online and offline sources for distribution and accessibility
on platforms like Twitter that limit posts to 280 characters (formerly 180).
In contrast, derivative information behavior consists of redistributing (e.g. retweeting)
generative and synthetic information, posting URLs and content copied from external sources, or
recommending information sources to other users, for instance by suggesting social media accounts
to follow. Derivative information accounts for most social media posted during a crisis. Across the
26 disasters examined by Olteanu et al. (2015), traditional and Internet media agencies post an
average of 42% of all disaster-related information, while remote users account for 38% (local
eyewitnesses account for the remaining 9%). However, suggesting the importance of social
networks for distributing information, studies observe that local users affected by a crisis are more
likely retweet the relatively-sparse generative information created by other local users than the bulk
of information distributed by the globally-distributed crowd (Kogan et al., 2015). Creating “a user-
38
driven cycle of shaping and re-shaping a shared interaction and information space” ” (Starbird et
al., 2010, p. 247), derivative information behaviors among crowds of social media users function
as information filters and recommendation systems that shape information access during a crisis
(Starbird et al., 2010, p. 247).
Information Disseminated on Social Media During Crises
Generative, synthetic, and derivative information behaviors among social media users
contribute to an information-rich information space during a disaster. Interestingly, the general
types of information citizens and officials post on social media remain remarkably similar across
natural and man-made disasters. Moreover, the type of information posted on social media changes
with each phases of crisis: preparation, mitigation, response, and recovery (Palen, Vieweg, Liu, &
Hughes, 2009; Wukich, 2016).
Across generative, synthetic, and derivative information, studies observe consistent types
and proportions of information posted during disasters. Olteanu et al. (2015) provide the most
comprehensive study by qualitatively coding Twitter posts collected across 26 natural and man-
made disasters. Table 3-1 describes the types of information content posted on social media during
a crisis, the average proportion of posts providing this information, and studies observing this type
of information content in the course of content analyses of crisis-related social media posts. The
findings of Olteanu and others provide a description of common information behaviors during
disasters, however, important features of information posted during emergencies, and the ability to
discover this information using multiple social media data collection methods, will be discussed in
Phase II of this study.
39
Table 3-1. Types of information posted on social media during crises.
Relevant, crisis-related Information (32%)
An admittedly “catchall category,” Olteanu et al. (2015) observe diverse social media reports including, for example, miscellaneous news reports and suspect information (i.e. Boston Bombing) Coordinating response (Muralidharan, Rasmussen, Patterson, & Shin, 2011; St Denis et al., 2012; Starbird & Palen, 2011; Takahashi, Tandoc, & Carmichael, 2015) Evacuation information (Vieweg et al., 2010); Coordinating relief efforts (Murthy & Gross, 2017; Takahashi et al., 2015) Government criticism & political action (Mazer et al., 2015; Murthy & Gross, 2017; Starbird et al., 2015; Takahashi et al., 2015) Animal management (Vieweg et al., 2010; White, Palen, & Anderson, 2014) Weather information , wind, visibility, flood levels, etc. (Grace, Halse, et al., 2019; Murthy & Gross, 2017; Vieweg et al., 2010)
Sympathy & Emotional Support (20%)
Memorializing victims (Takahashi et al., 2015) Thoughts and prayers (Mazer et al., 2015; Olteanu et al., 2015)
Affected individuals & Environmental Impacts (20%)
Requesting help (Purohit, Castillo, Imran, & Pandey, 2018; Takahashi et al., 2015) Reports of injuries, missing persons, & lives lost (Imran, Elbassuoni, Castillo, Diaz, & Meier, 2013b; Mazer et al., 2015) Reconnect family members (Takahashi, Tandoc & Carmichael, 2015) Environmental impacts (Starbird et al., 2015)
Donations and Volunteering (10%)
Monetary donations (Imran, Elbassuoni, Castillo, Diaz, & Meier, 2013a) Digital volunteering information (Purohit et al., 2014; Starbird & Palen, 2011, 2013; Vieweg et al., 2010)
Caution and Advice (10%)
Messages providing safety tips, travel recommendations, etc. (Imran et al., 2013a; Olteanu et al., 2015; Vieweg et al., 2010)
Infrastructure and Utilities (7%)
Damage reports (e.g. road blockages) and service disruptions (e.g. electricity outage) (Grace, Halse, et al., 2019; Imran et al., 2013a; LaLone et al., 2017; Tien, Aibek, Benas, & C. Pu, 2016; Vieweg et al., 2010)
Using Information Disseminated on Social Media During Crises
The information citizens, crisis response officials, and media organizations disseminate on
social media becomes, in turn, resources for situated actions among citizens and response officials
during a crisis (L. Palen et al., 2009; Leysia Palen et al., 2010; Leysia Palen & Liu, 2007). In this
40
regard, social media can contribute to “situational awareness” and, in turn, inform decision-making
in crisis situations (Vieweg et al., 2010). According to the common definition provided by Endsley
(1995), situational awareness describes a “state of knowledge” concerning the perception of
elements in the environment, comprehension of relations among elements, and projection of these
elements’ statuses in the future.
Social media can contribute to situational awareness as information resources describing,
and directing attention to (through derivative crowd behaviors) (Starbird & Palen, 2012), events
occurring in crisis-affected areas (e.g. damage to infrastructure, people in need of assistance,
resource locations, etc.) (Saleem et al., 2014; Sutton, Palen, & Shklovski, 2008), indicators and
warnings of events likely to occur as the crisis progresses (e.g. emergency warnings, public
opinions on evacuation orders, etc.) (Bean et al., 2015; Gow, Waidyanatha, & Anderson, 2007;
Sutton et al., 2014, 2015), and information coordinating the provision and delivery of humanitarian
aid (Purohit et al., 2014; Starbird & Kate, 2013; Starbird & Palen, 2011). For citizens, social media
also provides opportunities to access real-time, local information that may be unavailable from
traditional media sources during a crisis (Dailey & Starbird, 2017; Kogan et al., 2015; Shklovski
et al., 2008; Sutton et al., 2008).
Meanwhile, studies find crisis response officials adopt social media in emergency
situations (Gomez & Passerini, 2004; Hughes, & Palen, 2009; Y. Zhang et al., 2015). Among
government agencies and NGOs, social media often emerges as a source of information when
traditional media sources are unavailable or when existing emergency reporting channels (e.g. 911
or 1-1-2) break down or reach capacity (Kaewkitipong, Chen, & Ractham, 2012; Tapia & Moore,
2014; Tim, Pan, Ractham, & Kaewkitipong, 2017). In these circumstances, social media enables
crisis responder to communicate warnings, directives, and response updates to affected citizens
(Hughes, & Palen, 2009, 2012, Sutton et al., 2014, 2015), as well as monitor publicly-available
41
information to detect emergency events (Avvenuti, Cresci, Marchetti, Meletti, & Tesconi, 2014; Q.
Huang & Xiao, 2015; Tien et al., 2016; Xu et al., 2016), identify citizens’ questions and concerns
(St Denis et al., 2012), gauge community responses to crises (e.g. evacuation progress) (Cameron,
Power, Robinson, & Yin, 2012; Martín, Li, & Cutter, 2017), and gather sundry situational
information (Beneito-Montagut, Anson, Shaw, & Brewster, 2013; Chatfield & Brajawidaga, 2012;
Hong, Fu, Torrens, & Frias-Martinez, 2017; Kavanaugh et al., 2012; Mergel & Bretschneider,
2013).
Before crises, response organizations use social media to disseminate emergency warnings
to local citizens before emergencies (Houston et al., 2015; Rice & Spence, 2016; Veil, Buehner, &
Palenchar, 2011). During response and recovery operations, government agencies and NGOs use
social media to inform affected locals on the progress of relief operations, understand the
effectiveness of communication campaigns (Anson, Watson, Wadhwa, & Metz, 2017; Beneito-
Montagut et al., 2013; Martín et al., 2017), and coordinate relief efforts (Takahashi et al., 2015) by
communicating, for example, the location and availability of emergency supplies (Olteanu et al.,
2015; Tim et al., 2017). Lastly, during recovery, government agencies have turned to social media
to reconnect family members and memorialize victims (Olteanu et al., 2015; Takahashi et al.,
2015).
Coordinating Social Media Use During Disasters
Thus for over a decade, Crisis Informatics research has examined the use and impact of
Information and Communications Technologies (ICT) among people confronted with natural and
man-made disasters (Hagar, 2010). Within this research, studies of social media have addressed
how platforms such as Twitter and Facebook have decentralized information flows during disasters,
42
finding that effective uses of social media often become possible when remote, digital volunteers
coordinate information flows between affected citizens requesting assistance and officials
positioned to provide help. This “articulation space” in which digital volunteers work to coordinate
the situated actions of citizens and response officials remains a primary focus for Crisis Informatics
researchers who seek to enhance cooperation between citizens and officials during disasters
(Hughes & Tapia, 2015).
Within this articulation space, digital volunteers use social media to coordinate the
response activities of citizens and officials in two ways: distribution and monitoring. Social Media
Distribution (SMD) involves distributing and redistributing information resources (e.g. warnings,
procedures, updates, etc.) that shape the accessibility and visibility of information and guide the
situated actions of citizens during a crisis (Chauhan & Hughes, 2017; Denef et al., 2013; Kogan &
Marina, 2016; Starbird & Palen, 2011; Starbird et al., 2010; Sutton et al., 2014). In contrast, Social
Media Monitoring (SMM) involves collecting and analyzing information citizens post on social
media to curate reports that can support crisis responders’ situational awareness and decision-
making (Cobb et al., 2014; Hughes & Shah, 2016; Starbird & Kate, 2013). Crisis responders have
outsourced this work to remotely-located digital volunteers, sometimes organized as Volunteer
Operations Support Teams (VOST), who provide officials with needed technical and human
capacities to cope with the volume and variety of social media data generated during a crisis and
supplement the limited and overburdened staff of crisis response organizations (Cobb et al., 2014;
St Denis et al., 2012).
As the relationship between digital volunteers and crisis responders evolves, crisis response
organizations increasingly institutionalize workflows around open-source platforms and
information anticipated from digital volunteer communities (Soden & Palen, 2016). Taken
together, the use of social media during a crisis involves a) distributed and interdependent
43
activities— or cooperation that b) require articulation work of (re-)distribution and monitoring,
often performed by digital volunteers, to align— or coordinate— the distributed activities of
officials and citizens engaged in disaster response (Schmidt & Simonee, 1996).
By focusing on large-scale disasters, however, empirical contexts for Crisis Informatics
research have repeatedly shifted with the eruption of each new crisis: earthquakes in Haiti (2010)
and Nepal (2015), the tsunami in Japan (2011), and Hurricanes Sandy (2012) and Harvey (2017)
in the United States (Kogan et al., 2015; Reuter & Kaufhold, 2018; Samuels, Taylor, &
Mohammadi, 2018; Soden & Palen, 2016). As a result, researchers have paid less attention to the
articulation space coordinating social media use in geographic communities during periods of
stability nevertheless beset by routine incidents and intermittent emergencies that local citizens and
emergency responders encounter and cooperatively manage as a part of community life. Here
research in Crisis Informatics and Community Informatics overlap, framing a context for research
that examines community emergency response.
Thus, while the articulation space of digital volunteers working to distribute and monitor
social media in disaster has been often examined (Cobb et al., 2014; St Denis et al., 2012; Starbird
& Kate, 2013), the ongoing workings of the analogous articulation space in communities during
periods of stability, the context of community emergency response, remains relatively unexamined.
In contrast to crisis responders, prior research on community social media use finds that public
safety officials often lack sufficient staff, tools, policies, and trust to make effective use of social
media at all (Hiltz et al., 2014; McCormick, 2016; Plotnick et al., 2015; Reuter et al., 2016). This
following section describes these barriers to recognize a gap in current literature regarding the
objectives guiding current or prospective social media use during periods of stability, and the
cooperative activities, sociotechnical resources, and associated coordinative work for SMD and
SMM required for communities to meet these public safety objectives.
44
Barriers to Social Media Use in Community Emergency Response
Prior studies in Crisis and Community Informatics focus on emergency managers (Hiltz et
al., 2014; Hiltz & Plotnick, 2013, 2016; McCormick, 2016; Plotnick et al., 2015; Reuter et al.,
2016; Shan et al., 2017) and PIOs (Hughes, 2014; Hughes, & Palen, 2012; Hughes, & Shah, 2016)
supporting Emergency Management Agencies as critical adopters of social media and social data
analytics for SMM and SMD. These studies find that most officials perceive utility in the adoption
of social media for distributing public information and monitoring information distributed on social
media among people in the community for awareness of developing events, rumors, and reports of
emergency or calls for assistance (Hiltz & Plotnick, 2016). Among community emergency
managers, Plotnick et al. (2015) observe the use of social media for: “public alerting or reassurance,
public relations, monitoring special events, increasing situational awareness, providing specific
information to the public, countering rumors, sharing information with other organizations, and
sharing information on behalf of partners” (p. 188).
However, despite general perceptions of utility and extensive use in some jurisdictions,
studies commonly find that most officials face multiple barriers to the adoption and use of social
media. Officials report a lack of time, skills, and trained staff to make effective use of social media
such as Twitter for SMD and SMM, as well as barriers in the form of absent or restrictive policy
guidelines, lack of information processing tools to prevent information overload, and, most
commonly, general distrust of information posted on social media platforms (Hiltz et al., 2014;
Hiltz & Plotnick, 2013, 2016; McCormick, 2016; Plotnick et al., 2015; Y. Zhang et al., 2015).
These barriers are described in further detail below.
45
Lack of Staff
Emergency managers frequently point to a lack of staff and available time as reasons why
they cannot utilize social media for either SMD or SMM (Hiltz et al., 2014; McCormick, 2016;
Plotnick et al., 2015; Y. Zhang et al., 2015). In a survey of over 200 emergency managers, Plotnick
et al. (2015) found that only half made any use of social media, with a lack of staff and time most
often reported as the primary obstacle to adoption and use. As Zhang et al. (2015) observed among
county-level emergency managers preparing for Hurricane Sandy:
limited social media usage at this stage was due to the fact that staff, resources, and first
responders were doing all they could to stay on top of the more traditional forms of
communication. Whether or not local officials had the capability or intention of utilizing
social media, it was a common theme among respondents that social media information
acquisition was not a priority in the immediacy of the storm. (p. 4)
Since municipal governments and their agencies, including EMAs, typically operate within
significant financial and personnel constraints, opportunities for emergency managers to
experiment in the adoption and use of social media remain limited. Overlapping with a lack of staff,
and largely as a consequence of little time and resources for experimentation, EMA staff often lack
experience and training to effectively utilize SMD as a supplement to the Integrated Public Alert
and Warning Systems (IPAWS) that can be implemented alongside traditional emergency
communications strategies (Hiltz et al., 2014; Hiltz & Plotnick, 2016). Moreover, a lack of
experience, skills, and financial resources restrict EMAs from developing information processing
tools that would enable effective uses of SMM.
Lack of Tools
The volume, velocity, and variety of social media data generated within communities and
the global crowd of social media users provides additional challenges to the adoption of SMM
46
among emergency managers and PIOs (Bruns & Liang, 2012; Cobb et al., 2014; Hiltz & Plotnick,
2016; Hughes, 2014; Hughes, & Shah, 2016). In particular, the volume of social media data creates
challenges for officials seeking to monitor social media for information that can support situational
awareness and decision-making (Cobb et al., 2014). A lack of technical tools to prevent information
overload, or the inability of humans to effectively make sense of large quantities of data, persists
among emergency management agencies (Hiltz & Plotnick, 2013). Despite needs for technologies
to assist resource-strapped agencies collect and filter social media data, technical tools remain
scarce or unaffordable for most community EMAs (Hiltz & Plotnick, 2016; Plotnick et al., 2015).
In addition to technologies that can collect, process, and curate or visualize social media
data (Y. Zhang et al., 2015), officials require the necessary technical and practical expertise for
SMM that remain limited due to the lack of EMA staff, time, and training opportunities (Hiltz &
Plotnick, 2016; Plotnick et al., 2015). Forms of expertise involve, for instance, the ability to
effectively manage cross-medium public information campaigns, or integrate information obtained
from social media within established emergency response and management operations. As Plotnick
et al. (2015) describe: “We find that emergency managers lack practical guidance as to how to
judge [social media], evaluate it, categorize it and make it useful. Because of a lack of
understanding and experience, response organizations offer blanket rejections of social media.” (p.
191). The lack of expertise and experience overlap with the lack of personnel and staffing resources
common to community EMAs, and thus prove major obstacles to adoption and use. However, it
should be noted that these obstacles pervade local emergency management and not simply the most
resource and capacity constrained agencies. In the survey conducted by Plotnick and colleagues,
few differences were observed between counties of varying population size. Thus, as a result, and
in lieu of widespread adoption and use beyond simple, periodic information dissemination,
knowledge of what constitutes effective expertise for SMM or the functional requirements for
47
technologies supporting SMM remain scarce in the context of community emergency response and
management (Cobb et al., 2014; Hughes, & Shah, 2016).
Lack of Policy
Moreover, emergency management agencies lack formal policies, guidelines, or accepted
best practices for social media use (Beneito-Montagut et al., 2013; Hiltz et al., 2014; McCormick,
2016; Plotnick et al., 2015). This reflects a common situation for many government agencies that
are adopting social media as a component of both their informal and, slowly, formal practice
(Mergel & Bretschneider, 2013). Approached according to Mergel and Bretschneider’s (2013)
three-stage adoption process for social media use among government agencies- experimentation,
emergent standard operating procedures, and institutionalization- significant diversity exists among
emergency managers who variously report having no policy, or having informal or, in the minority
of cases, formal policies related to the collection and dissemination of social media (Plotnick et al.,
2015; Y. Zhang et al., 2015).
A significant barrier in these indications of nascent policy development are institutional
policies prohibiting social media use altogether. In Plotnick et al.’s (2015) survey, fully one-quarter
of U.S. emergency managers who reported having a formal policy, had a policy completely
prohibiting the official use of social media (p. 186). Moreover, the narrow existence of a use policy
must not be confused with effective or comprehensive guidelines for social media for dissemination
strategies or monitoring functions (Houston et al., 2015; X. Lin, Spence, Sellnow, & Lachlan,
2016). Additionally, studies observe that both formal policy and informal, standard practices among
emergency managers tends to develop only with successive emergency response experiences of
adoption and use (Mergel & Bretschneider, 2013; Plotnick et al., 2015; Y. Zhang et al., 2015). This
48
would suggest that barriers to adoption involving a lack of resources and capacities also stand to
delay the policies that might otherwise encourage adoption and, when in use, utility.
The lack of staff, tools, and policies restricting EMAs from employing social media has
opened possibilities for public participation in community emergency management. This
participation often involves training and deploying “trusted volunteers,” such as those organized in
VOSTs, to assist EMA operations during crisis (St Denis et al., 2012). More open and distributed
crowdsourcing efforts to assist officials in monitoring information posted on social media remain
possibilities (Bruns, Burgess, Crawford, & Shaw, 2012; L. Palen et al., 2009), with speculation that
social media crowds might provide an initial layer of vetting to filter social media posts and mitigate
information overload among response officials (McCormick, 2016). As Plotnick et al. (2015)
describe, “when data is collected, it is not seen as trustworthy enough to use directly, but rather is
seen as worthwhile as a first step (monitoring, situational awareness) in gathering the data needed
for action.” (p. 188). Such hesitancy points to perceptions of utility in the use of social media as
well as the need for integrating and verifying information obtained from social media within the
existing, and typically closed, workflows of public safety officials
Lack of Trust
Lastly, the pervasive distrust of information posted on social media by the public remains
the most insistent barrier to social media adoption and use among public safety officials (Beneito-
Montagut et al., 2013; Hiltz et al., 2014; Hiltz & Plotnick, 2016; Tapia, Bajpai, Jansen, & Yen,
2011). In the experience of emergency managers she interviewed, McCormick (2016) reports that
the “most common and important challenge was difficulty verifying data collected through
informal channels and the subjectivity that characterized this data, making it difficult to use
49
correctly” (p. 216). While public safety officials do monitor social media for information in certain
situations (Dailey & Starbird, 2017; Tapia & Moore, 2014), Hiltz and Plotnick (2016) find that the
lack of trust surrounding information reported on social media represents “a barrier that…
prevent[s] the use of social media for information collection at this time” (p. 267). Specifically, this
lack of trust arises over the credibility of people posting information on social media and the
accuracy of information they post. As McCormick (2016) observed, this perceived lack of
credibility and accuracy creates the challenge of verifying information reported on social media
before it can be incorporated in decision-making processes.
Credibility represents the trust public safety officials place in the source of information. As
Tapia and Moore (2014) describe:
The concept of trust here is trust in a person, trust in an organization and trust in a network,
all of which produce data that can be seen as more accurate because of the human agents
involved. Trust in the network of emergency responders becomes an advance filtering
system of data, culling and categorizing social media data. (p. 504)
In this respect, challenges immediately appear in the use of social media as a source of information
for emergency response, citizens, remain outside the centralized communication channels and their
vetting mechanisms, specifically the 911 system and 911 call takers and dispatchers who mediate
and translate information conveyed from the public to appropriate first responders. Emergency
managers often remain an addition step removed from this information, contacted by emergency
dispatchers or emergency response officials at an incident command post only if an emergency
requires additional resources and inter-agency coordination for response.
Distrust in the accuracy of information posted on social media by the public represents a
common concern among emergency managers (Hiltz & Plotnick, 2016; McCormick, 2016) and
PIOs (Hughes, 2014; Hughes, & Palen, 2012), echoing similar concerns expressed by crisis
responders (Anson et al., 2017; Tapia et al., 2011; Tapia, Moore, & Johnson, 2013; Tapia & Moore,
50
2014). The perception of social media as prone to rumor, factual inaccuracy, and exaggeration,
coupled to its informal composition, lead officials to be dismissive of its professional utility.
A common concern among emergency managers involves the apparent “subjectivity” of
posted social media. That is, the accuracy of information, especially when looked to as reports of
“on the ground,” situational conditions, is perceived to be easily distorted by the affective state or
ignorance of citizens. As McCormick (2016) describes, “subjectivity makes it difficult for
responders to use, especially when attempting to understand the severity of an event or prioritizing
calls for clean-up and recovery” (p. 218). In this regard, distrust extends from the use of imprecise
language in social media (Saleem et al., 2014; Vieweg et al., 2010). Evaluating social media posted
during a flood, including a post describing a local park as “completely underwater,” Saleem, Xu,
and Ruths (2014) conclude that “interpretive issues such as this created by imprecise descriptions
of flooding damage, in combination with photos taken from different perspectives (and even of
spatially distinct parts) of a location made proper assessment of flooding progress exceedingly
difficult” (p. 162).
Research seeking to facilitate trust in social media has, for example, identified
“trustworthy” content features as a basis for automated filtering techniques (Halse, Tapia,
Squicciarini, & Caragea, 2018). In contrast, Tapia and Moore (2014) find that officials utilize
information on social media when it is provided by trusted sources, observing that “trust in people
trumps trust in information” (p. 508). In this regard, trust rests on the credibility of the information
source: “trust in a person, trust in an organization and trust in a network, all of which produce data
that can be seen as more accurate because of the human agents involve” (p. 504). This finds support
in the results of a survey conducted by Hiltz et al. (2014) where the most commonly indicated
technological aids included the ability to filter twitter streams by source (user) category and display
the geographic location using a GIS (p. 606). The reported motivation for filtering among the
51
emergency managers surveyed would be to stay aware of information posted by other, related
officials or credible organizations (e.g. Red Cross). However, an important corollary to concerns
for credibility and accuracy among officials is noted by Dailey and Starbird (2017), who observe
that officials will utilize information reported on social media when it has been corroborated with
information obtained from trusted sources.
Table 3-2. Phase I research questions.
What objectives, cooperative tasks, and coordinative practices characterize social media use in
community emergency response?
RQ1 What objectives guide emergency responders’ uses of social media?
RQ2 What cooperative tasks and resources characterize emergency responders’ uses of
social media?
RQ3 What coordinative practices characterize social media use in community emergency
response?
Social Media Use as Cooperative and Coordinative Work
Barriers to the use of social media among community public safety officials- staffing
constraints, a lack of information processing tools, prohibitive or immature policy guidelines, and
distrust in the credibility of social media users and the veracity of information they post on social
media, continue to impede social media use among community emergency management and
response officials seek during periods of stability.
On one hand, these barriers resemble those crisis responders encounter during disaster but
have relied on the articulation work of remote, digital volunteers to overcome, raising the
possibility that community officials might similarly engage local, “community volunteers” to
coordinate the use of social media in their communities. On the other hand, officials working within
52
local incident command systems encounter difficulties when asked to cooperate with citizen
volunteers acting outside the organizational and procedural structures that centralize authority and
organize accountability among public safety professionals (Hughes & Palen, 2012; Hughes, 2014).
As a result, these existing barriers to social media use might be differently approached as
breakdowns in coordination between the distributed, ad hoc activities of local citizens using social
media platforms and the operations of officials working within incident management systems. This
reinterpretation, however, requires addressing gaps in the literature regarding the objectives,
activities, and sociotechnical resources characterizing social media use in community emergency
response.
In this regard, social media use in community emergency response can be considered from
the vantage of Computer-Supported Cooperative Work (CSCW), specifically through Kjeld
Schmidt’s understanding of cooperative work and articulation work. These interrelated concepts
provide an opportunity to reinterpret observed barriers to social media use- the lack of staff, tools,
and trust among community emergency responders as breakdowns in coordination, (and the
possibility for cooperation), that constrain the effective use of social media during an emergency.
According to Schmidt and colleagues (Schmidt, 1994, 2008; Schmidt & Bannon, 1992;
Schmidt & Simonee, 1996), cooperative work represents a particular form of human activity
characterized by interdependent, distributed activities of people working toward a common goal.
“Cooperative work arrangements” develop among people engaged in separate (distributed in time
and space) but mutually dependent activities. People within a cooperative work arrangement share
a “common field of work” consisting of the objects each uses and modifies for her activities that,
because these objects are required in the activities of others, creates conditions of mutual
dependence among actors. More precisely, Schmidt (1994) defines cooperative work as
53
“interdependent work activities carried out in relation to a common field of work and mediated by
changes to the state of the field of work” (p. 17).
As activities mediated by Information and Communications Technologies (ICTs), social
media use in emergency response can be understood as a form of CSCW among people living
together in a community. That is, the activities of emergency response involve mutual dependencies
among distributed citizens and officials using social media to accomplish public safety objectives
(Schmidt & Simonee, 1996). A citizen calling 911 for medical assistance, for example, depends on
the capacities and resources of dispatched paramedics while, at the same time, paramedics depend
on citizens to request medical assistance.
The distributed activities of a cooperative work arrangement, however, demand
coordination, what Schmidt refers to as articulation work: “activities arising from the fact that the
work requires and involves multiple agents whose individual activities need to be coordinated,
scheduled, meshed, integrated etc., in short, articulated” (Schmidt, 1994, p. 17). The mutual
dependencies characterizing cooperative uses of social media in emergency response require
articulation work to coordinate the distributed, yet mutually dependent, activities of citizens and
officials (Schmidt & Simonee, 1996): for paramedics to receive a call for help depends on 911
communications infrastructure, and the work of 911 dispatchers who obtain and forward priority
information from the caller to paramedic. Emergencies present unique opportunities for the use of
social media in emergency response work. These opportunities, in turn, create unique coordination
requirements to achieve community public safety objectives (Gurstein, 2005, 2007).
To understand the coordinative requirements posed by a cooperative work arrangement,
Schmidt draws on Strauss (Strauss, 1985) to outline “elemental objects” of articulation work: tasks
(what is the objective for cooperation?), activities (how can the objective be accomplished?), and
roles (who is available, and who will perform what activities?). Coordinative requirements also
54
arise for the “common field of work.” These include the conceptual, informational, technical
resources required by the actors within a cooperative work arrangement (Schmidt & Simonee,
1996).
Research Questions
Drawing on Schmidt’s theorization of cooperative and coordinative work, phase one of this
study addresses the following questions: (RQ1) What objectives guide emergency responders’ use
of social media? While social media distribution and social media monitoring have been generally
understood to support situated action and situational awareness among citizens and emergency
responders (Hiltz & Plotnick, 2016; Leysia Palen et al., 2010; Leysia Palen & Liu, 2007; Reuter et
al., 2016; Vieweg et al., 2010), respectively, the tasks that social media distribution and social
media monitoring might be deployed to accomplish remain poorly understood in the context of
community emergency response.
(RQ2) What cooperative tasks and resources characterize emergency responders’ uses of
social media? Considering the objectives guiding the employment of social media, this study
examines the cooperative arrangements that emerge among citizens and officials using social media
by decomposing the distributed but mutually-dependent activities constituting these arrangements,
and identifies the sociotechnical resources employed in these activities.
(RQ3) What coordinative practices characterize social media use in community emergency
response? Understanding the coordinative requirements for social media distribution and
monitoring, we examine how communities might (re-)align the use of social media around available
actors, (including community volunteers), and sociotechnical resources to overcome barriers of
staffing, tools and information overload, and trust they now frequently encounter. The realignment
55
of activities through coordination mechanisms available in communities (Schmidt & Simonee,
1996), provides opportunities to develop stable community infrastructures that can enhance the
ability of a community to prepare for, respond to, and recover from more-or-less routine
emergencies and, potentially, large-scale disasters (Carroll, Shih, & Kropczynski, 2004; Dailey &
Starbird, 2017; Mark & Semaan, 2008; National Institute of Standards and Technology, 2015;
Semaan & Hemsley, 2015; Semaan & Mark, 2011).
The development of community infrastructures recalls FEMA’s “Whole Community”
concept, and the development of sociotechnical infrastructures as a central aspect of resilience-
building: “existing structures and relationships that are present in the daily lives of individuals,
families, businesses, and organizations before an incident occurs can be leveraged and empowered
to act effectively during and after a disaster strikes” (FEMA, 2011, p. 5). While community
emergency response involves a unique cooperative context of community work and requires
deliberate forms of community coordination supporting social media use, realigning community
actors and resources for emergency response across successive emergencies stands to organize
capacities and resources that can be deployed in the event of a large-scale crisis or disaster (Aldrich
& Meyer, 2015; Amir & Kant, 2017; Manyena, 2006; Murphy, 2007; Ospina, 2014; Tagliacozzo
& Arcidiacono, 2015). In our discussion, we consider how the realignments for social media use in
emergency response can organize infrastructures contributing to community resilience.
Method: Scenario-Based Interviews
Interviews involve researchers speaking with people and “exploring the ways in which
subjects experience and understand their world. It provides a unique access to the lived world of
56
the subjects who in their own words describe their activities, experiences, and opinions” (Kvale,
2008, p. 9). Interviews also provide researchers with a method of data collection. In this phase of
research scenarios are used to facilitate and systematize data collected from community emergency
management and response officials through conversations that contribute to the interpretation and
analysis of their lived world. The following outlines the scenario-based interview research design-
including the process of scenario construction and participant recruitment, interview and data
collection procedures, and method of analysis the interview data. Lastly, this section describes
evaluative criteria for qualitative research.
Scenario-based Interview Design: Scenario Construction
Interviews were conducted around use-case scenarios, “short stories about hypothetical
characters in specified circumstances, to whose situation the interviewee is invited to respond”
(Finch, 1987). Using scenarios prompted public safety officials to consider possibilities of social
media use involving community volunteers in concrete situations of use (Carroll, 2015). Scenarios
were constructed as part of the interview research design process, which concerns recruiting
participants and developing interview protocols that focus data collection on the research questions
to be explored during this first phase of investigation (Kvale, 2008, p. 35).
To construct each scenario, we began with actual situations of social media use described
by officials serving a Class 3 county that was the site of an early, pilot study, and extrapolated these
accounts by integrating activities and technologies observed among digital volunteers in disaster
response (Cobb et al., 2014; St Denis et al., 2012; Starbird & Palen, 2011). As an exploratory study,
we iteratively modified the three constructed scenarios across our interviews with community
57
officials to make each both plausible to interviewees and critical from a design perspective (Carroll,
2000). As Carroll (2015) describes, such scenarios can guide a process of theory-based design:
Instead of grounding scenarios in direct observation and empirical literature- and in so
doing tying them to the present and the past- we could ask merely that the scenarios be
intelligible and plausible to domains actors, with respect to what happens or could happen
and what matters or could matter in domains, and to be productive and critical to domain
analysts and designers, with respect to raising intellectual and practical challenges and
opportunities, and scientific hypotheses. (p. 195)
Each scenario was constructed and modified to facilitate insight into the purposes, cooperative
arrangements, resources, and associated coordinative requirements of social media use in
community emergency response and management. In this sense, we sought to use the constructed
scenarios as a form of probe, “a design-oriented way to acquire inspirational glimpses of
communities targeted for design” (Boehner, Vertesi, Sengers, & Dourish, 2007).
The three scenarios begin with the prompt: “The following scenarios are set in fictional
Laurel County. They imagine how an established group of community volunteers - the "Laurel
Watch" (LW) - might support an emergency manager by monitoring and disseminating information
on social media.” Each scenario (1-3) then proceeds with a statement introducing a potential
emergency event: Scenario 1. A rally for a national political candidate will be held in Laurel county;
2. Rumors are spreading of contamination to the county water supply; 3. Weather forecasts warn
of a severe winter storm.
During pilot interviews with officials in a class 3 county, participants discussed an occasion
where a candidate for national political office visit the county. For the visit emergency management
officials activated a “social media component” which included volunteers who actively monitored
social media to ascertain the exact time of the candidate’s arrival as well as locations where crowds
were assembling in the county. Using this example, Scenario 1 describes the work of community
volunteers during a political rally held in Laurel county:
58
LW volunteers monitoring social media report the planning of a protest among both locals
and people from around the state. You notify and forward a report to the police chief. On
the day of the rally, volunteers curate important social media posts for you, including: posts
detailing traffic jams surrounding the event; pictures at the rally that show and describe the
developing protest; and campaign posts suggesting that the candidate- and the rally- will
be delayed multiple hours. Traffic congestion and police efforts to manage the rival
demonstrations have led to significant delays on the main road leading to the rally. You
ask LW volunteers to identify and use relevant hashtags to retweet official traffic and safety
advisories.
Interviewed officials sought to use social media as prior rallies held by the candidate had motivated
rival demonstrations and, in some cases, violence between opposing supporters. Scenario 1
describes such events to prompt participants to consider the value of social media monitoring to
identify and mitigate hazards surrounding large demonstrations and unpredictable crowd behavior.
In contrast to providing reports of on-the-ground events, Scenario 2 considers a rumor
sparked and quickly disseminated on social media claiming contamination to the county’s water
supply:
Volunteers notify you of a rumor gaining traction on social media that the local reservoir
has been contaminated. Posts point to an apparent report of trace pollutants detected in a
stream adjacent to a local factory feeding into the reservoir. You call the local water
authority to understand the situation. The accounts disseminating the rumor appear to be
associated with people living outside the state yet are generating discussion among
residents. You instruct volunteers to disseminate official statements dispelling the rumor,
point out their source, and direct citizens to alternative sources of information.
Numerous participants identified the use of social media to spread rumors and expressed interest
in quickly detecting and correcting misinformation being disseminated in their communities.
Scenario 2 prompts officials to consider the utility of community volunteers in performing this
function.
Lastly, scenario 3 imagines a severe weather storm- a common hazard experienced by
emergency response and management officials in Pennsylvania:
Before the storm, volunteers disseminate weather warnings, emergency and non-
emergency contact information, and preparation advice on social media. On a map you see
the locations of curated social media posts reporting power outages, snowfall conditions,
and pictures of fallen trees. One area seems especially hard hit. One volunteer monitoring
59
social media, also a member of a community service organization, indicates the prevalence
of assisted-living persons in the affected area who might require help.
The particulars of the scenario originated with experiences of participants interviewed during the
pilot study but were found to be common among public safety officials across the state. The
scenario prompts officials to again consider the utility of community volunteers using social media
to disseminate and monitor community information. The inclusion of hazards to assisted-living
citizens reflects the common concern of emergency management and volunteer organizations for
vulnerable populations (e.g. the elderly) during power outages that cause loss of heating during
winter storms.
Following Carroll (Carroll, 2000; Carroll & Kellogg, 1989), we approach each of the use-
case scenarios as a designed artifact embodying theories of users, technologies, and activities of
use. In this sense, use-case scenarios describing cooperative social media distribution and
monitoring constitute theoretical artifacts that can be evaluated (and falsified) by domain experts,
in this case, public safety officials. Interviews provide opportunities for evaluating the embodied
theories of social media use, with the scenarios serving as the bases for semi-structured interviews
focusing on the objectives, activities, and sociotechnical resources embodied in the scenarios
(Schmidt & Simonee, 1996).
Participant Recruitment
As this phase of research focuses on barriers to social media use encountered among
community officials, interviews were sought with U.S. county-level public safety officials,
especially emergency managers, who coordinate incident and emergency management operations
among local government agencies (e.g. EMA, PSAP, emergency services), volunteer organizations
(e.g. VOADs), and the public within a geographic jurisdiction. We began by soliciting participants
60
serving EMAs in the state of Pennsylvania to arrange phone or in-person interviews. Altogether we
interviewed 30 participants representing 16 county EMAs, including 14 EMA directors, 6 EMA
planning and operations managers, five PIOs, and five PSAP supervisors (Table 3-3).
Table 3-3. Interviewed public safety officials by county population.
County Class Participants Total
1 (>1.5m) - -
2 (>500k) P11, 29-30 3
3 (>210k) P5, 12, 13, 16, 17, 21, 27, 28 8
4 (>145k) P4, 6-8, 18, 19, 22-26 11
5 (>90k) P3 1
6 (>45) P1, 2, 10, 14, 15, 20 6
7 (>20) - -
8 (<20) P9 1
30
The populations served by these officials vary widely. In Pennsylvania, counties are
classified from 1-8 based on population size, with Class 1 and 2 counties located in urban centers
with populations over 1.5m and 500k respectively, while Class 8 counties feature less than 20k
people living in very rural areas (Table 3-3). In their survey of U.S. county-level emergency
managers, Plotnick and Hiltz find “few differences in social media use or perceived barriers
associated with county characteristics of population size or urban vs. rural composition,” to include
the size of EMA staff (Hiltz & Plotnick, 2016). However, as we explain in the following analysis,
we only observed officials of more populous jurisdictions (Class 2-3) employing SMM.
Participant recruitment involved three sources and proceeded through snowball
recruitment methods. Utilizing contacts existing from previous research, emergency management
and response officials in the local community were recruited for interviews. However, to better
understand general practices among emergency management officials, directories from the
61
Pennsylvania Emergency Management Agency (PEMA) listing county and other local emergency
management officials were utilized to recruit interview participants. While conducting interviews
developed from these sources, subsequent participants were recruited through participant referrals
(i.e. snowball sampling). The latter became especially important as the exploratory nature of the
study, and the subsequent findings, focused on PSAP supervisors and their experiences of social
media use.
Interview Protocol and Data Collection
The scenarios and accompanying interview protocol represent a design-probe (Boehner et
al., 2007) and instruments for collecting data (Creswell, 2009, p. 175) (Appendix A). Consistent
with the exploratory nature of the study, semi-structured interviews are distinguished by their
limited and open structure, sufficient to orient discussion toward the phenomena of study, and open-
ended to allow in-depth exploration of emergent topics (Williamson, 2013, p. 361). Furthermore,
as with the iterative construction of use-case scenarios, the exploratory nature of the study calls for
the refinement of the interview protocol as research process progresses. This refinement occurs in
the interplay between stages of interviewing and analysis, allows for emergent themes to develop
over the course of the study (Glaser & Strauss, 1967; Kvale, 2008, p. 43). Smith and Osborne
(2008) prescribe general guidelines for semi-structured interviews to include building a rapport,
flexible ordering of questions, exploring the interviewee’s interests and concerns, and using probes
to explore areas of interest. In this respect, the scenarios and interview protocol represent flexible
guides for conducting each interview and probes providing an entry-point to open discussion
according to the participants’ reflections, interests and concerns regarding the use of social media
and community volunteers in emergency response and management.
62
Before each phone interview, written scenarios were forwarded to participants by email.
These served as the basis for exploratory, semi-structured interviews that followed emergent topics
participants brought up as they reflected on the situations, actors, and activities described in each
scenario. Interview data was collected either through written notes or audio recordings that were
later transcribed and analyzed. In the case of this study, some interviews will be audio recorded
while others, to accommodate the wishes or comfort of people interviewed, will not. In cases where
interviews are not audio recorded, written notes are taken to document important themes and, at
times, direct quotes. In accord with best practices for qualitative research and the conduct of semi-
structured interviews, the study was submitted to and approved by Pennsylvania State University’s
Institutional Review Board (Appendix B). The terms of this approval stipulate that the data will be
appropriately stored and destroyed after a period of five years.
Scenario-based Interview Data Analysis
The third stage of interview-based research involves analysis of data collected through
interviewing participants and then evaluating this analysis (Kvale, 2008, p. 101). Analysis begins
during the interview process and proceeds to formal methods of analysis using the written notes
and, when available, transcripts of conducted interviews. Following Schmidt and Simonee (1996)
we analyzed emergency managers’ discussions of these scenarios by coding and analyzing the
tasks, activities and roles, and resources defining cooperative work arrangements of SMD/M, as
well as the sociotechnical resources (conceptual, informational, and technical) that support the
common field of work of these arrangements.
To analyze the interview data, a combination of selective and open coding techniques will
be used. Selective coding begins with a priori conceptual categories that are then applied to data in
63
analysis (Glaser, 1978). Selective coding will categorize sections of data collected through
interviews as relating to the articulation of cooperative work arrangements (actors, roles, activities,
tasks), or the resources engaged in practices of cooperative work (information, material, procedural,
conceptual) (Kreps, 1978; Schmidt, 1994, pp. 25–26). Importantly, selective and open coding
overlap and proceed simultaneously, that is, each unit of analysis will be coded as either a direct or
indirect report, and an additional code that emerges through the open coding process that will be
described below.
Open coding involves a grounded, inductive process by which data categories are
developed and articulated as an interpretive, explanatory framework (Bryant & Charmaz, 2007;
Glaser & Strauss, 1967). Open coding begins with the development of conceptual categories held
to interpretively account for the data. According to Glaser and Strauss (1967) concepts (i.e.
categories) “should be analytic- sufficiently generalized to designate characteristics of concrete
entities, not the entities themselves. They should also be sensitizing- yield a “meaningful” picture”
(p. 28-9). The initial stage of open coding finds the development of as many categories as emerge
through the data.
Open coding proceeds through a process of constant comparison by which developed
categories are compared to new data to either subsume newly encountered within existing
categories or recognize occasions of “breakdown” where extant conceptual framework does not
apply (Agar, 1986; Trauth, Quesenberry, & Huang, 2009). At these points, the coherence of the
extant conceptual framework fail, and new categories called for and developed to account for
emergent themes in the data (Trauth & Jessup, 2000). This comparative process results in the
iterative development of conceptual categories that, importantly, become increasingly articulated
and defined with breakdowns and emergent category development.
64
Lastly, at a point when breakdowns and consequent emergent themes no longer arise in the
process of analysis (Lee & Baskerville, 2003), the coding process ends and the conceptual
framework is presented as a theoretical account of the phenomena under investigation. The
importantly the conceptual categories do not represent discrete descriptions but must be described
through their inter-definitions with the other categories developed (Glaser & Strauss, 1967, p. 114).
Stern (2007) states this as a central evaluative condition of inductive, qualitative analysis:
“Integration of the finished product needs to be executed in such a way that every component is in
harmony with every other component” (p. 114). The developed framework, consisting of a set of
interrelated categories and their associated properties describing the activities of social media use
in community emergency response and management, can then be written and presented to readers
as the final stage of the interview process (Kvale, 2008).
Evaluative Criteria for Qualitative Research
Evaluation of qualitative research involves a negotiation among researchers and those
evaluating their research, whether they be academic reviewers or the members of a community in
which the research is concerned (Easterby-Smith, Golden-Biddle, & Locke, 2008; Guba & Lincoln,
1989). This negotiation rests largely on how the representations of research findings are explained,
and concerns the procedural description of the whole process of investigation, from problem
formulation though data collection, analysis, and the articulation of findings (Creswell, 2009, p.
190). As such, evaluative criteria include the systematic description of the procedures of research,
as well as general evaluative procedures undertaken within the process of research itself. The latter
include contextualization, triangulation, and member checking (Guba & Lincoln, 1989; H. K. Klein
& Myers, 1999).
65
Contextualization “requires that the subject matter be set in its social and historical context
so that the intended audience can see how the current situation under investigation emerged” (H.
K. Klein & Myers, 1999, p. 73). Contextualization thereby concerns the contingency of the
phenomena investigated, a contingency continually in a process of change and toward which
process the research and object of design will contribute (however modestly). As Klein and Myers
(1999) write:
When the researcher does field research, the results of his or her work are influenced by
the total history of the organization and the research itself becomes a part of the
organization's future history. The principle of contextualization requires that this be
explicitly reflected. (p. 73-4)
In this study contextualization is developed through the thick description of research that begins by
addressing prior literature, focuses on current phenomena by describing interview findings, and, in
the discussion chapter, addresses the conditions, limitations and implications of this research. These
discussions address the contextual nature of investigation and the idiosyncratic characteristics of
community emergency response and management as practiced in State College, Pennsylvania,
Pennsylvania at large, and, in Phase 3, Charleston, South Carolina. This thick description attending
to the relationship between the historical, situated, and reflexive character of the investigation
contributes to the contextualization of research to reflect awareness of its positioning and
contribution within a “total history.”
Triangulation “refers to checking inferences drawn from one set of data sources with those
collected from others” (Golafshani, 1990; Seale, 1999). This study seeks comparisons among
different sources data collected and analyzed during each phase of investigation, across phases, and
between the findings of this study and extant literature. During Phase I, findings from interviews
are compared across interviewees. Furthermore, the use-case scenarios generated in one interview
serve as resource and probe for others, including the series of workshops, allowing for the
progressive comparison and evaluation. Triangulation is also achieved by situating the current
66
study in the landscape of extant literature on emergency response and management and response
and crisis and risk communication related to social media. Findings of Phase I inform and are
compared to those of Phases II and III. During Phase II, the types of information observed during
the case examined are compared with those observed in other research across diverse crises and
geographic areas. Phase I directly motivates Phase III, with the latter examining in detail the
activities of detection, verification, and integration identified in Phase I.
Lastly, member checking consists of “data, analytic categories, interpretations, and
conclusions are tested with members of those stakeholding groups from whom the data were
originally collected...” (Guba & Lincoln, 1989, p. 314) and constitutes an important procedural
strategy for interpretive research (Miles & Huberman, 1984). Extending the triangulation of
findings across the phases of investigation, this study member checks the validity of the findings
of Phase I and II by “testing” them with officials in Phase III.
Recalling again Carroll’s (2015) description of the evaluative criteria for scenario-based
design, that scenarios be “be intelligible and plausible to domains actors…” and “productive and
critical to domain analysts and designers,” Phase III calls for domain actors to evaluate and use the
design products of Phase I and II. The findings that extent from the design workshops conducted
during Phase III provide opportunity to check the intelligibility and plausibility of previous findings
and develop further insights that prove productive and critical when understanding the distributed
sensemaking process of integrating social media within emergency response, and the associated
design requirements for systems that facilitate this integration. This accords with the procedural
form of validity central to qualitative research (Creswell, 2009, p. 190), and Golden-Biddle &
Locke’s (1993) evaluative criteria of authenticity, plausibility and criticality which look to research
that convincingly grasps the phenomena investigated, makes a valid contribution to issues of shared
concern, and expands scope for insight and inquiry.
67
Findings: Coordinating Social Media Use in Community Emergency Response
Conducting scenario-based interviews to describe the objectives, activities, and
sociotechnical resources of social media use for emergency response and management, this study
explores possibilities for community volunteers to distribute and monitor information in ways that
coordinate uses of social media that make communities more resilient. Across the interviews, public
safety officials discuss social media as either of two sociotechnical systems in which citizens,
officials, and prospective volunteers are assigned clearly-defined roles. First, and principally,
officials use social media as a mass notification system for distributing timely messages to the
public to coordinate the public response to an emergency. Here, Social Media Distribution (SMD)
requires aligning officials and citizens via community influencers, citizens whose social networks
include large followings of local citizens and, as a result, can volunteer as effective redistributors
of official messages.
Second, and deployed only in urban, populous jurisdictions, some officials approach social
media as an early warning and indirect reporting system that supplements existing, direct
emergency reporting (i.e. 911) by using social data to discover early warnings of incidents and
additional, situational information surrounding emergencies. For these officials, Social Media
Monitoring (SMM) offers a new way to coordinate emergency responses within incident
management systems by enabling a proactive approach to emergency dispatch that seeks to
anticipate both the direct report and occurrence of emergency. Below we describe the objectives,
activities, and resources constituting these sociotechnical systems to discuss the forms of
coordination they each require
As a result, this study finds that coordinating social media use requires aligning community
social media networks with local incident management systems. Specifically, SMD requires
aligning publics and public safety officials via “community influencers” who can effectively
68
redistribute community information. SMM requires aligning citizens posting indirect emergency
reports with first responders capable of providing aid via Public-Safety Answering Points, 911 call
centers with unique infrastructures for verifying and integrating crowdsourced information. These
alignments suggest limited roles for community volunteers, notably redistributing official
information, but, at the same time, stand to reshape incident management systems by pushing
officials to proactively process both direct and indirect emergency reports and develop protocols
that can facilitate collaborative sensemaking among distributed citizens and officials.
By reinterpreting barriers resulting from a lack of staff, policy, tools, and trust in social
media as breakdowns in community coordination, this study identifies opportunities to overcome
these barriers by realigning roles and resources among community citizens and emergency response
and management officials. In doing so, the findings of this study describe an information
infrastructure supporting emergency response that differs from infrastructures supporting remote
social media use during disasters. This chapter presents the findings of this phase of investigation
regarding the objectives, cooperative activities, and resources of SMD and SMM in community
emergency response and, in turn, motivates the studies conducted in Phase Two and Phase Three.
Social Media Distribution
As prior research observes (Hiltz & Plotnick, 2016), the public safety officials we
interviewed (n=25) use social media primarily to distribute information to the public. Officials
approach social media as a mass broadcast system to issue the public authoritative, up-to-date, and
directive information. In this respect, officials regard social media as one component of a suite of
broadcasting systems, including the Emergency Alert Service (EAS) for broadcasting messages
69
over television and radio, Wireless Emergency Alerts (WEA) for broadcasts to geolocated mobile
devices, and subscription-based services for sending email, SMS, and mobile app notifications.
Regarding the objective of SMD, officials attempt to coordinate public action before,
during, and after emergencies. When an EMA posts on Twitter, for example, “Avoid X road. Take
detour Y” officials attempt to coordinate the activity of on-site responders managing a car accident,
with the activities of citizens taking (hopefully) the advised detour on their morning commute.
SMD therefore involves sending the right message, to the right people, who can then take the right
actions: “What it comes down to is being able to get them the information when they need it, make
it valuable to them to make decisions, and be self-sufficient for a certain period of time” (P13).
This involves aligning three interdependent activities among citizens and officials: gatekeeping,
redistribution, and public response (Figure 3-1).
Figure 3-1. Social media distribution information flow from officials to citizens.
Gatekeeping, or the process by which municipal agencies control what information reaches the
public (Hughes & Palen, 2012), guides officials’ use of social media and constrains the possible
70
roles for community volunteers in SMD. Scenario three begins by suggesting volunteers might
distribute messages to the public:
Scenario 3a. Before the storm, volunteers disseminate weather warnings, emergency and
non-emergency contact information, and preparation advice on social media.
Every official we interviewed rejected this idea out of hand: “We can’t have outside people putting
out information for us that has not gone through our vetting process” (P14). This gatekeeping
process involves coordinating activities often distributed among staff across municipal agencies—
information gathering, verification, inter-agency coordination, and message approval— to control
what official information reaches the public. Conversely, public action involves the situated actions
of citizens using official information. This information and the activities they inform can, of course,
vary widely: severe weather notifications, traffic advisories, preparation advice (e.g. storm kits),
etc. Crucially, however, the possibility of public action hinges on aligning official gatekeepers with
citizens who stand to use official information.
As gatekeepers, EMA staff are responsible for disseminating accurate, safety-critical
information on behalf of the municipal government. The task of disseminating this information to
the public therefore remains closely controlled within an internal process detailing how each
message should be composed and disseminated, to include obtaining specific approvals at the
agency and county levels. This process enables accountability and coordination within the EMA
and across government agencies in a jurisdiction.
This leads officials to describe SMD with respect to two domains of message control. The
first encompasses the internal processes and control emergency managers exercise over the official
messages they put out to the public. As another emergency managers explains, not only must this
role- official gatekeeper- remain reserved to EMA staff, but beyond vetting, selecting, and sending
information, their control over social media as a mass broadcast system effectively stops:
71
If we would put something on our social media, on Facebook and Twitter, and they take
that and retweet it or tell their neighbors, of course we have no control over that. But for
us to actively, solicit them to do that for us, I don’t see us doing that. (P18)
While prior research often observes the official gatekeeping role assumed among emergency
responders (Hughes, & Palen, 2012), the imaginary of emergency managers finds this role
accompanied by the desire to act as community gatekeepers. That is, emergency managers’ shape
their use of social media in ways that attempt to manage the visibility and accessibility of public
information circulating among the community.
EMA staff, for instance, selectively share information with the public to ensure accuracy
and protect sensitive information (e.g. victim’s names), but also to anticipate and manage what
information should be visible to the public. As one emergency manager rues:
One of the other problems that arises with social media and postings from a governmental
entity, is that the second you post it invariably somebody... forwards your post along with
a misleading assessment or statement and that’s the one that gains traction and once that is
done you have no control over that. (P12)
In the desire to control what information gains traction in the community the gatekeeping role
assumed by emergency managers extends to encompass the management of public information
outside the control of officials and official protocols. Information selection as well as re-posting
and cross-posting authoritative information becomes an important method for extending this
control:
If you are doing something very simple, say following up a media release where you
reiterate to those who did not see the media [release] saying- “hey, the following roads are
closed within [state] county because of flooding”- then that’s a benefit. But if you’re giving
a status update to the response of a given organization, there is always somebody who
thinks they know better than the on-scene responders as to what action is necessary. And
so, it becomes a viral comment… (P12)
Selecting and restricting what information reaches the public can articulate grounds for
coordinating the community response (i.e. detouring traffic) by mitigating the risk of coopted or
alternative messages gaining visibility. The work of reiterating by retweeting, reposting, and
72
linking to other, official sources can mitigate against the cooption of official messages, while
supporting emergency managers’ role of aligning multi-agency (EMS, fire, police, etc.) activities
with those of the public.
The threat of misinformation and rumor therefore represents a serious challenge to
emergency managers’ role as community gatekeepers. Scenario 2 introduces a rumor of water
contamination:
Scenario 2b. The accounts disseminating the rumor appear to be associated with people
living outside the state yet are generating discussion among residents. You instruct
volunteers to disseminate official statements dispelling the rumor, point out their source,
and direct citizens to alternative sources of information.
As community gatekeepers, emergency managers seek to monitor social media for rumors,
misinformation, and expressions of uncertainty and fear that, they hope, can be corrected or
assuaged by disseminating official information. Rumor control represents the most common social
media monitoring activity among emergency managers we interviewed. As one emergency
manager explains:
We look to see what the general population is saying about, whether it be power outages,
or what’s really happening, and so on and so forth, and if the perception in the public is
wrong then naturally we put out correct information so that information could be corrected.
(P11)
Rumor control, as an aspect of External Affairs, also constitutes one of sixteen functions performed
by an emergency operations center (EOC) according to the National Incident Management System
(NIMS), the policy framework for federal, state and local emergency management activities in the
United States (DHS, 2008). When a municipal EOC is activated and staffed during an emergency,
“operators track calls, locate previously unknown pockets of damage and identify misperceptions
that the PIO should try to dispel” (PEMA, 2009). Though not explicitly concerning the use of social
media, rumor control represents a unique instance in which institutional policy motivates social
media use.
73
Redistribution
After releasing information to the public, officials’ control over SMD- and the role of
gatekeeper- effectively ends: “If we would put something on our social media, on Facebook and
Twitter, and they take that and retweet it or tell their neighbors, of course we have no control over
that” (P18). Here the articulation space of SMD emerges around the redistribution activities (i.e.
retweeting and sharing) of community influencers- volunteers who stand to coordinate information
flows between officials and citizens. However, the features of this articulation space remain
obscure: officials know neither the citizens with whom they are communicating or the community
influencers who can coordinate this communication.
Scenario one describes a local emergency manager directing the distribution of event-
related information with volunteers of the “Laurel Watch” during a political rally:
Scenario 1c. Traffic congestion and police efforts to manage the rival demonstrations have
led to significant delays on the main road leading to the rally. You ask LW volunteers to
identify and use relevant hashtags to retweet official traffic and safety advisories.
As the scenario illustrates, reaching the right people (e.g. people at the rally) with the right message
(e.g. detours to take), likely requires deliberate coordination. For many officials (n=9), SMD can
be used to reach more or different demographics of citizens than possible with existing distribution
channels. Considering the role of redistribution, these officials readily acknowledge the role of
citizens: “we depend on, at that point… people to go ahead and grab hold of that information and
continue to push it out” (P7). However, when considering the role community volunteers might
play in redistributing official messages, as suggested in scenario one, emergency managers and
PIOs often highlight experiences coordinating press releases with local television and radio
stations. Officials readily describe the “great relationships” they develop with local media in the
community, to whom they can call directly to arrange public notifications (P9). Though seemingly
analogous to these traditional activities of PIO work, most officials found the idea of working with
74
community influencers- citizens who can voluntarily redistribute official messages to a large or
specific group of people in the community- relatively foreign.
While most officials we interviewed actively utilize SMD (n=25), these officials typically
neither know who uses social media in the community or take deliberate efforts to reach them.
Paradoxically, these officials employ SMD to quickly reach demographics in their communities.
Through SMD, for instance, officials attempt to communicate with the people who do not subscribe
to the county’s mass notification system: “even with this new system [Everbridge] we have
probably less than 5% of the county enrolled in it” (P13). For other officials, SMD provides access
to unique audiences: “kids who don’t even use email anymore” (P18). “We utilize it for reaching
that demographic, because it’s a certain demographic that utilizes social media...,” explains another,
“it is usually the younger generation that is constantly on it, and that is our way of reaching them”
(P10).
At the same time, emergency managers understand SMD as a poor way to reach other
populations in the community. Among these are people who may be especially vulnerable during
emergencies:
A lot of times the people that need assistance are not the ones on social media because you
don’t find a lot of the elderly folks on Twitter, you don’t find people with access and
functional needs on Twitter, and economically depressed people cannot go out and buy a
$600 iPhone… (P11)
Beyond anecdotal evidence, however, officials we interviewed rely on assumed notions of the
“imaginary audiences” who use social media in their jurisdictions (Dailey & Starbird, 2017;
Marwick & boyd, 2011). Importantly, these assumptions motivate the adoption and use of social
media among public safety officials as they attempt new ways to reach segments of the population
in their community.
75
Social Media Monitoring
What objectives motivate officials to monitor information citizens post to social media?
Prior research describes social media as a general source for situational awareness information
(Vieweg et al., 2010), however, our scenarios sought further insight into the objectives that
determine motivations and requirements for SMM in community emergency response and
management. Scenario one, for instance, reads:
Scenario 1b. On the day of the rally, volunteers curate important social media posts for
you, including: posts detailing traffic jams surrounding the event; pictures at the rally that
show and describe the developing protest; and campaign posts suggesting that the
candidate- and the rally- will be delayed multiple hours.
While emergency managers we interviewed appreciated the general usefulness of situational
information— “if people have information like that or tell us they are going to send us photos that
is fantastic because the more information that we can get the better we can help somebody” (P15)—
officials (n=12) often expressed difficulty imagining how community volunteers could support the
routine operations of incident and emergency management. Some expressed interest in deploying
SMM to identify rumors circulating in the community (n=4), or, in the event of emergency, for
rapid damage assessment using information (especially images) citizens post (P7, 24).
However, for officials with actual SMM experience (n=8), SMM offers a distributed,
indirect reporting system that can be used to coordinate incident and emergency dispatch by
providing early warning of incidents before they are reported to 911: “We know that people will
share things on social media and, in some cases, put things up before calling 911 or calling the
power company” (P13). Moreover, as multiple people often report information on the same
incident, social media provides a distributed source of situational information that can supplement
sparse or incomplete information obtained from individual 911 callers and on-site emergency
responders. Officials find citizens using social media alongside existing reporting channels; people
76
will call 911 to report incidents they observe on social media (P3), or “put something on Twitter or
Facebook and assume that someone has already called it in” (P18). Across our interviews with
these officials, we identified six mutually-dependent activities of SMM that must be coordinated
among citizens and officials: reporting, detection, verification, integration, dispatch, and response.
Reporting, Dispatch, and Response
Deploying SMM would seem to mirror the cooperative roles already assumed by citizens
and officials for incident management: citizens report incidents, officials dispatch the appropriate
emergency services who respond with on-site fire suppression, medical assistance, security, etc.
However, as an indirect incident reporting system, the activities of reporting and dispatch radically
change.
First, rather than wait for citizens to directly report an emergency by calling 911, officials
seek to proactively detect and dispatch first responders to early warnings of incidents by monitoring
indirect reports- information citizens post on social media but do not deliberately communicate to
officials. “We are essentially utilizing a resource that we don’t have direct access to for information
gathering,” explains the emergency manager of a Class 3 county:
Right now, we are dependent on somebody else initiating the share of information with us,
whether it’s a responder in the street using a radio, a civilian calling 911 or a posted point
of contact number, unless someone sees something and says something there is a delay in
us gaining the information. Every second that we are unaware of an event or situation
delays, correspondingly, any response that we can provide to whoever is in need in
whatever capacity. (P12)
Second, and as a result, responsibility for coordinating SMM shifts from emergency managers or
PIOs- officials prior research has pointed to as critical adopters- to the staff of Public-Safety
Answering Points (PSAPs). Furthermore, deploying SMM within a PSAP involves three new
activities- detection, verification, and integration- that coordinate the flow of information between
77
citizens posting indirect reports and officials who rely on information PSAPs dispatch (Figure 3-
2).
Figure 3-2. Social media monitoring information flow from citizens to officials.
Detection
Deploying SMM within PSAPs remains, to our best understanding, extremely rare. Only
two emergency managers and one PSAP supervisor interviewed made use of this arrangement due
to the unique organizational alignments present in their communities. Both emergency managers
were directors of Public Safety Departments in urban, class three counties that administratively
unified the county EMA and PSAP. In most jurisdictions, the EMA and PSAP operate along distinct
administrative and operational divisions roughly corresponding to emergency and incident
management, respectively.
However, adopting SMM within a PSAP, rather than an EMA, presents distinct
advantages. First, the demands placed on EMA staff to continuously monitor social media has
78
posed a persistent barrier to the use of social media. Adopting SMM/D within PSAPs, however,
overcomes this barrier, proving “absolutely beneficial for the emergency management side here, of
course during a disaster we are here, but the [911] dispatch is there around the clock, 24/7” (P18).
Unlike EMA offices, and most Emergency Operations Centers (EOCs), PSAPs are continuously
staffed and stand to provide around-the-clock monitoring and distribution.
Second, as the 911 dispatch centers already coordinates information flows information
between citizens and public safety officials, the PSAP possesses the unique sociotechnical
infrastructure for verifying and integrating indirect reports of incidents obtained from social media
with direct reports obtained from 911. In most communities, the EMA remains removed from 911
information flows: “When those two operations are split [EMA and PSAP], at least in my
experience, what I take for granted as information coming out of the 911 center, there is also
sometimes a significant delay in that knowledge rising to the level of EMA operations.” (P12).
Again, this delay reflects the different operational mandates of the two agencies, with PSAPs
responsible for dispatching information to first responders surrounding more-or-less routine
incidents, and EMAs concerned with emergency operations planning and response.
Deploying SMM within a PSAP, however, creates new challenges. Foremost is the
required shift in posture from reactively processing direct reports of incidents received via 911, to
proactively detecting indirect reports of incidents on social media. The officials who have
encountered this challenge recognize the need for technologies that can access hyperlocal social
media data (data created by people within a geographic community) to detect early warnings of
incidents: “The idea was to mine information using keywords for things like power outages, flooded
[areas], road closures, that kind of thing would really benefit us greatly,” explained one official
(P13). One PSAP utilized Geofeedia, a commercial platform for collecting and mapping geolocated
information, to monitor social media alongside 911 calls:
79
We would have it running in the background of the 911 center, and across Twitter and
Facebook we had a list of approaching 75-100 keywords…surprisingly a lot of times social
media gets posted on an event, particularly a fire or a law enforcement occurrence, and it
would hit social media before a 911 call was made. (P12)
Use of Geofeedia abruptly halted in October 2016, when Twitter, Facebook, and Instagram denied
the platform commercial access to data after reports were published describing its use by law
enforcement agencies during the 2014 and 2015 protests in Ferguson, Missouri and Baltimore,
Maryland, respectively (Cagle, 2016). As the unified EMA/PSAP director explained, without
access to hyperlocal data the platform became “no use to us” (P12). Lacking a replacement, the
PSAP no longer monitors social media.
Verification and Integration
As officials do not immediately trust information citizens post on social media, once an
indirect report is detected it must be verified for accuracy. The officials we interviewed sought to
verify reports of incidents by following, often informal, protocols for corroborating information
obtained from social media with information obtained from trusted sources, either on-site officials
and 911 callers, or, in some cases, other social media users. Importantly, PSAPs provide the
sociotechnical infrastructure for these processes of verification. “When we get the information
[from social media], we will… vet it and investigate…,” one emergency manager explains:
Let’s say they [911 dispatcher] gets something that someone puts out [on social media], “I
just drove by John Doe warehouse and I see smoke and flames shooting out of the roof.”
They get that information and, hopefully that person has also called 911, but what they
would do is let the supervisor know, they would probably yell it in the room, “hey, I just
saw this.” They would then send, law enforcement-wise, someone there to verify as soon
as possible that there was an issue there (P18).
As infrastructure, the PSAP draws together, in the same room, the personnel and ICTs needed to
verify incidents reported on social media. These include dispatchers who monitor for incidents
80
detected on social media, supervisors authorized to dispatch officials to investigate incident reports,
as well as the information systems (i.e. radios, CAD) for receiving 911 calls and communicating
with on-site first emergency responders to obtain (non-)corroborating information. The other
unified emergency manager/PSAP director observes the same:
If we get a report that there are shots fired, and nobody can confirm it, generally most
municipalities will send an actual patrol out to check the area, and once they’re on the
street, because we are a unified operation here- we handle 911, emergency management,
and emergency dispatch- there is no entity in [Pennsylvania] County that we don’t have
direct, two-way communications with through a secured radio system. So once an officer,
patrolman, fireman, regardless, is asked to check a report that has not been verified, we
rely on our two-way radio communication… (P12)
This unique sociotechnical infrastructure for detecting and verifying incidents reported on social
media enables the PSAP to coordinate the situated actions of citizens and officials surrounding
emergencies.
Occasions arise, however, when officials act on social media information before it can be
corroborated and verified using official sources. In the case of multiple, original citizen reports or
visual evidence of an incident on social media, officials might immediately dispatch aid before
verifying the accuracy of the information. In these cases, officials are willing to corroborate a
reported incident on social media using only other- untrusted- social media sources:
I’m going to try and validate the source. Is this the only one [tweet] that says this or is there
someone else saying it too? I’m going to look to see if people are just retweeting the same
thing that I just saw, or are they getting it from someone else? Or, are they all the originators
of this information? (P8)
In the hope of dispatching a response more quickly, for certain incidents officials are willing to
substitute untrusted “crowd” sources for trusted sources of information that can provide
provisionally verify of the accuracy of detected, indirect reports. “If you get one report you’ll
probably send one person out there to check it,” describes another emergency manager, “If you get
ten people, you might start rolling out the fire company” (P18).
81
These descriptions unpack officials’ understanding of trustworthy information reported on
social media by distinguishing between needs for accurate and useful information. Narrating a
hypothetical dialogue between a 911 caller and emergency dispatcher, an official describes these
needs:
“Oh, I think there's a real bad accident or fire here...”
“Well do you think there's injuries?”
“Well it looks pretty bad” … So now you dispatch for injuries and you get there and it's a
flat tire or something stupid like that. It happens all the time.” (P24)
Officials require reports on social media to be accurate or factual (e.g. that there is, indeed, an
accident) and useful for that information to be trustworthy: information suitable for dispatching
emergency response personnel and resources (e.g. an ambulance and fire engine are not required
for a flat tire). With respect to the usefulness of information, pictures stand to better communicate
the necessary attributes of an incident than text alone:
Pictures tell me more than people. The problem is human beings speak the language of
affection. I need the language of fact. I need facts. Okay, so we talk about good/bad,
hot/cold- that means nothing to me in decision-making. I need it's raining an inch an hour,
it's 70 degrees, the building has collapsed, flames are coming out of 5-6 windows- I mean,
I need the facts. “It’s a big fire.” Ok, well what is that? Well to you that could be the shed,
I don’t know. To me a room and contents bedroom fire is not a big fire. So a... well taken
picture is worth much more than a human being’s description. (P24)
While crowd measures and posted images or video can provide officials with a measure of
information accuracy (e.g. is there really a fire?), social media content provides useful information
to the extent its content includes indicators or attributes dispatchers require to classify an incident
for emergency dispatch (e.g. what address/coordinates? type/size of fire?). This classification, in
turn, determines the response personnel and resources dispatched to the incident.
Lastly, and importantly, officials deploy SMM to supplement existing information flows,
requiring the integration of indirect reports obtained from social media, with direct reports obtained
from 911. Here PSAP staff draw upon existing protocols for 911 call taking that specify information
requirements and their relative priority when gathering information from 911 callers:
82
Our primary thing is to not gather as much information from additional sources as we can.
Our primary mission is to get as much information from the caller as you can, and as soon
as you have the location and incident type, dispatch help, and continue to take information
and push it to the help en route through a mobile data terminal. So, by the time an officer
or firefighter pulls up, if there is anything that would change their posture, upon entry to
that event, they have that information before they get there. (P12)
As supplemental information, officials utilize information reported on social media to address
information requirements not met by 911 callers, and vice versa. Moreover, the motivation of a
distributed early warning system suggests that useful SMM technologies would support existing
protocols of emergency dispatch by prioritizing incident alerts, consisting of incident type and
location, for the rapid dispatch of emergency services, and, second, the curation of situational
reports that provide relevant situational awareness information for use by emergency responders
en route to an emergency.
The activity of integration distinguishes SMM in emergency response, and in the PSAP
particularly, from most all other activities of social media use previously examined in Crisis
Informatics research. The necessity of information integration as a precondition for supporting first
responders’ situational awareness during emergencies requires incorporating social media data
within the already data-rich and data-heterogenous environment of Public-Safety answering Points
(PSAPs) and, more broadly, incident management systems coordinating emergency dispatch and
response. Where prior studies consider the processing of social media in largely data-homogenous
contexts, such as in the work of digital volunteers, incorporating social media within emergency
response requires process internal to community incident management systems. This requires
systems that can collect and filter social media data, and methods for analyzing and integrating
multiple, incomplete reports of information among multiple people and multiple data sources,
including 911 caller data, traffic data provided by platforms such as Waze, and various data
obtained from on-scene first responders, media sources, etc. This work requires information
integration to address issues of redundancy among heterogeneous data to identify and dispatch
83
unique, actionable information to first responders. It becomes useless to monitor, detect and verify
information reported on social media if that information is already known among emergency
response officials (for instance, if already reported by 911 callers), or cannot be integrated with and
therefore improve the accuracy and usefulness of existing information obtained using existing data
sources.
Discussion
Whereas experiences of large-scale disaster see cooperative uses of social media among
crisis responders on the ground, remote digital volunteers, and citizens requesting and providing
assistance, multiple barriers impede uses of social media among community emergency responders.
Across a series of scenario-based interviews, emergency responders articulate the objectives,
cooperative activities, and resources for social media use in community emergency response. These
interviews suggest that the lack of staff, tools, and trust impeding social media use among
emergency responders can be overcome by coordinating social media monitoring within emergency
dispatch centers, Public Safety-Answering Points (PSAPs) with unique infrastructures for verifying
and integrating citizen-reported information, and building relationships with “community
influencers,” citizens well-positioned in community networks to redistribute officials’ messages on
social media
This study provides new perspectives on barriers previously observed to prevent social
media use in community emergency response. Contrasting research focusing on emergency
managers as the primary adopters of social media analytics (Hiltz et al., 2014; Hiltz & Plotnick,
2016; Shan et al., 2017), interviews with community emergency responders point to emergency
dispatch staff serving in PSAPs as critical adopters. Incorporating social media monitoring into
84
emergency dispatch operations enables the integration of information from social media,
emergency callers, and first responders, opening opportunities to enhance situational awareness
among emergency responders and citizens (via social media distribution) during emergencies. Yet
for emergency dispatch staff to proactively process direct and indirect emergency reports will
require new analytic software and procedures that can facilitate distributed sensemaking among
citizens, emergency dispatch staff, and first responders.
Furthermore, observing the need to develop ties between emergency response agencies and
citizens and community organizations well-positioned to re-distribute official messages on social
media suggests updates to PIO practices traditionally concerned with building cooperative
relationships with local television and radio stations (Hughes, & Palen, 2012). Looking to
“community influencers” as emergency communication channels available to emergency
responders presents new opportunities for citizens to participate in public safety that can bolster the
social capital of communities during emergencies (FEMA, 2011; Grace et al., 2019; Murphy,
2007). Overall, by identifying cooperative arrangements that sustain effective uses of social media,
this study points to community infrastructures that can help emergency responders overcome
barriers impeding the use of social media in emergency response.
The Right Staff for the Right Task
Our interviews find that simply adding more public safety staff remains insufficient for
officials to use social media effectively: social media distribution and monitoring each require
alignments among community citizens and officials each positioned vis-à-vis unique sociotechnical
infrastructures. It is insufficient, for example, to assign a full-time PIO to monitor social media, if
that information can only be verified, integrated, and dispatched by PSAP staff. Similarly, the
85
availability of an emergency manager to distribute official notices on social media remains
insufficient if community influencers, mobilizing community networks, do not redistribute these
messages to reach their multiple, intended publics.
Overcoming staffing barriers requires aligning the specific activities of SMD/M among
community actors who are not only available and capable to take on social media duties, but
appropriately positioned with respect to the cooperative arrangements and sociotechnical
infrastructures required for coordinating these tasks. As prior research on barriers to community
social media use focus on the field of emergency management rather than emergency response, and
on emergency managers and Public Information Officers (PIOs) serving in Emergency
Management Agencies (EMAs) as critical adopters of social data analytics (Hiltz et al., 2014; Hiltz
& Plotnick, 2016; Hughes, 2014; Hughes, & Palen, 2012; Hughes, & Shah, 2016; Plotnick et al.,
2015), these studies neglect social media use among public safety organizations primarily
responsible for emergency response: emergency medical, fire, and police services and emergency
dispatchers serving in Public-Safety Answering Points (PSAPs).
Consequently, and in contrast to prior research in Crisis Informatics research (Hiltz &
Plotnick, 2016; McCormick, 2016), our interviews suggest that among community officials the
critical adopters and infrastructure for SMD/M are not emergency managers and the EMA, but
emergency dispatch staff and the PSAP, respectively. While EMAs lack staff to regularly
disseminate and monitor social media, PSAPs remain continuously staffed and resourced to verify,
integrate, and dispatch information reported on social media to first responders, or communicate
timely messages to the public.
Similarly, the role of community volunteers remains constrained. For SMD we find
immediate opportunities for community volunteers, and specifically those with large social
networks, to redistribute messages to the public. While gatekeeping functions will likely remain
86
reserved to officials alone, coordinating the redistribution of official messages in a community
stands to address what Gurstein (2005) refers to as the “last mile warning system,” where
community influencers constitute “the missing links—the last mile—from the “professional” early
warning system that governments can do best… to the “effective use” of the output of those systems
by local communities for early warning” (p. 15).
In contrast, the coordinative requirements for SMM, and specifically for early warning and
awareness for incident response, likely preclude community volunteers from performing the same
activities- monitoring, curating, and summarizing- that remote, digital volunteers now perform on
behalf of crisis response organizations during disaster. Rather, deploying SMM for incident
response requires PSAPs to perform activities of detection, verification, and integration, and
citizens to report incidents using both direct and indirect channels. The requirement that SMM
activities be conducted in hubs already coordinating emergency response operations also suggests
that these activities can take place at a distance (Dailey & Starbird, 2017), relying on remote,
continuously-manned state and regional Emergency Operations Centers (EOCs) and data
processing centers. To manage the Pennsylvania Turnpike, for instance, a state-level Traffic
Operations Center manages early warning systems for detecting and responding to traffic incidents.
Importantly, the center functions through a unified incident command system providing
coordinating for all agencies with responsibilities related to the turnpike (Pennsylvania Turnpike
Commission, 2018).
Tools for Information Access
While prior research has described a lack of tools allowing public safety officials to manage
high volumes of social media data (Hiltz & Plotnick, 2013, 2016), our interviews reveal that
87
barriers of information access often precede those of information overload. Officials we
interviewed mostly sought abilities to collect hyperlocal data originating from their geographic
jurisdiction. Moreover, our interviews find that officials find effective use of rudimentary filtering
methods (e.g. search terms) but often struggled to access enough local information for SMM to
prove useful. As the case of Geofeedia illustrates, access to social media data remains controversial,
commercially-fenced, and tightly moderated by social media platforms, with some providing more
opportunities for access (e.g. Twitter) than others (e.g. Facebook).
Thus, research pursuing the development of event detection and classification methods
(Imran, Castillo, Diaz, & Vieweg, 2015; Imran et al., 2013b, 2013a; Purohit et al., 2018; Rudra et
al., 2016; Vieweg, Castillo, & Imran, 2014) might be complimented with methods that can expand
access beyond, for example, the 1-3% of tweets geotagged within a geographic community
(Morstatter et al., 2013), as well as studies that examine the effectiveness of collection methods in
accessing situational information relevant to the context of CIEM rather than disaster management.
Grace et al. (Grace et al., 2017), for instance, infer Twitter accounts associated with a geographic
community on the basis of community-associated social network ties. Such geographic inference
methods could then be leveraged to collect hyperlocal social media data and by event detection
tools to identify and curate indirect reports of emergency posted in a community.
For SMD, officials also lack access to information identifying the “imagined audiences”
they hope to reach (Dailey & Starbird, 2017), as well as the community influencers through whom
they can reach them. Coordinating SMD first requires, then, access to information describing the
audiences or publics that form when large numbers of local citizens follow the account of a local
organization or person on social media, the latter can include local newspapers, politicians, and
celebrities. Identifying such community influencers marks potential volunteers who stand to align
publics with officials communicating public safety information. To coordinate the timely
88
redistribution of messages from officials to citizens vis-a-vis such volunteers, applications enabling
the automatic redistributing posts by official accounts (under specified conditions) could be
adopted by community influencers. Just as officials have traditionally developed relationships with
local broadcast media to distribute press releases, relationships could be developed with community
influencers to automatically redistribute official social media posts in emergency situations.
Trust in Protocol
Lastly, our findings provide further perspective on how public safety officials come to trust
information reported on social media. Rather than characteristics of information content (Halse et
al., 2018), or the source of information (Tapia & Moore, 2014), we find trust to be based in protocol,
such that information reported on social media becomes trustworthy when officials follow
procedures for verification, integration, and dispatch. Trust in protocol is highlighted in PSAP use
of SMM, where incidents reported on social media initiate processes of verification, integration,
and dispatch without preconditions for trust in the content or source of information.
Our interviews reveal that the accuracy of information reported on social media can be
verified when corroborated with trusted sources and, under certain conditions, information posted
by unknown, untrusted social media users. Crucially, whether trusted or alternative sources of
corroborating information suffice for verifying and, in turn, integrating information reported on
social media in activities of emergency dispatch remain situated determinations guided by formal
or informal protocols gradually developing among PSAP officials. These protocols coordinate
activities of detection, verification, and integration among emergency dispatchers and supervisors
as they, in turn, coordinate information flows between citizens posting indirect emergency reports
on social media and first responders receiving radio dispatches.
89
Importantly, the protocols 911 telecommunicators follow when obtaining priority
information from 911 callers and integrating information from multiple 911 callers can inform the
design of tools and cooperative workflows that would coordinate SMM within PSAPs. By
specifying what information is important in a given emergency, these protocols can inform how
tools filter and classify indirect reports of incidents, summarize situational information across
multiple indirect reports, or measure the accuracy of information using other social media posts.
Moreover, while slowly emerging, these protocols describe new sensemaking processes that
develop around the integration of multiple sources of crowdsourced data (e.g. 911 and social media)
within distributed cooperative arrangements of citizen reporters, PSAP telecommunicators, and
dispatched first responders that are reshaping how communities respond to incidents and
emergencies.
The findings of this study also point to requirements for the design of SMM technologies
that can facilitate trust among officials by processing and filtering accurate and useful information.
In this respect, SMM technologies would facilitate protocol-organized processes. Thus, while
features of social media data associated with information accuracy and utility can and should be
examined to inform the development of automatic classification techniques suitable for filtering
and detecting emergencies within real-time social media data, these technologies will facilitate trust
to the extent that they make this information available to officials following protocols of emergency
dispatch and the integration of information across data sources.
Furthermore, protocols 911 dispatchers follow to verify information reported on social
media require time-critical procedures for investigating the accuracy and usefulness of information.
This suggests that SMM technologies will be required that facilitate, first, the curation of incident
reports that include useful information immediately required by emergency dispatchers for incident
classification and dispatch (i.e. incident location and type), as well as available crowd measures
90
indicating the accuracy of the information. Second, SMM technologies might curate situational
reports to provide emergency responders with additional, salient information determined by the
incident type and context of response.
Limitations and Future Work
Our interviews with public safety officials suggest communities can overcome barriers
impeding uses of social media by realigning distribution and monitoring roles among existing
community actors- including volunteers- able to coordinate the activities of community emergency
response and management. However, interviewing only public safety officials, rather than citizens,
limits our findings to a partial understanding of the tasks, activities, and resources that shape these
cooperative uses of social media and their coordination requirements. The tasks guiding social
media use, as community objectives, must necessarily be negotiated among diverse community
stakeholders. Here we seek to represent the understandings of public safety officials to all
community stakeholders (Carroll & Rosson, 2007) and thereby contribute to this negotiation by
mapping a space for continued dialogue surrounding the evolving requirements for coordinating
social media use in communities.
Extending the findings of previous research (Hiltz & Plotnick, 2016; Hughes, 2014;
McCormick, 2016; Shan et al., 2017), this study details the coordinative requirements for social
media use in community emergency response and management. In this context, and in contrast to
the information infrastructure organized around digital volunteer work during crises (Cobb et al.,
2014; Robinson, Maddock, & Starbird, 2015; St Denis et al., 2012; Starbird & Palen, 2011, 2013),
coordinative roles for community volunteers transform and become less apparent when officials
91
prioritize mechanisms for rapid and targeted information dissemination and event detection
systems that can be integrated into existing incident management systems.
However, by addressing social media use in community emergency response and
management as a form of cooperative, community work, our interviews also reinterpret barriers to
social media use previously observed among public safety officials- a lack of staff, tools, and trust
(Plotnick and Hiltz, 2016)- as breakdowns in community coordination. We suggest that these
barriers can be overcome by aligning social media use within existing incident management
systems. First, these findings recommend incorporating social media capabilities within Public
Service Answering Points (PSAP), 911 call centers that already process citizen-reported
information and coordinate inter-agency and public incident notification and emergency dispatch.
PSAPs are continually staffed and, via trusted protocols, can uniquely verify, integrate, and
dispatch information reported on social media using existing sources of official and citizen-reported
information. However, incorporating social media within PSAPs requires technologies for
detecting and processing indirect reports of emergency on social media and procedures for
integrating this information with that obtained from existing data sources. In this regard,
community use of social media hinges on the incorporating systems and procedures for the
activities of detection, verification, and integration in PSAPs. Studies to elicit design requirements
for these sociotechnical activities remain necessary. Phase Three of this investigation takes up this
research.
Second, as this study points to barriers of information access preceding those of
information overload, the extent to which SMM technologies can effectively process information
reported on social media necessarily depends on access to geolocated social media data created by
citizens within a geographic community. As this study finds, community emergency responders
lack systems to collect this data and, in addition, find limited utility in information obtained from
92
available data sources. This need reflects general restrictions on data use among social media
platforms and, in the case of Twitter, limited ability to access geolocated data (Carley et al., 2016;
Crampton et al., 2013; Morstatter et al., 2013; Shelton, Poorthuis, Graham, & Zook, 2014). Phase
Two addresses this need by introducing a novel data collection method to complement existing
methods of collecting geolocated social media data and evaluates the utility of these methods by
identifying actionable, indirect reports of infrastructure damage and service disruption in the
context of a severe weather emergency.
Conclusion
Whereas experiences of large-scale disaster see cooperative uses of social media among
crisis responders on the ground, remote digital volunteers, and citizens requesting and providing
assistance, multiple barriers impede uses of social media among community emergency responders.
Across a series of scenario-based interviews, emergency responders articulate the objectives,
cooperative activities, and resources for social media use in community emergency response. These
interviews suggest that the lack of staff, tools, and trust impeding social media use among
emergency responders can be overcome by coordinating social media monitoring within emergency
dispatch centers, Public Safety-Answering Points (PSAPs) with unique infrastructures for verifying
and integrating citizen-reported information, and building relationships with “community
influencers,” citizens well-positioned in community networks to redistribute officials’ messages on
social media.
Crisis Informatics research often focuses on social media use for crisis response and
management during large-scale disasters rather than on community emergency response and
management during intervening periods of relative stability. This research describes use contexts
93
of information and resource scarcity in which crisis responders use social media as a source for up-
to-date information when other information sources are absent or unavailable. Lacking staff and
technologies to collect and curate information reported on social media, crisis responders rely on
information infrastructures that develop around the coordinative work of digital volunteers and
technologies they assemble to facilitate information flows between people using social media to
request assistance or provide situational reports during disaster, and crisis responders positioned to
provide aid. Consequently, research efforts to improve crisis response have examined the
organization and workflows of digital volunteers, and developed tools that assist remote
information work.
In contrast, research examining social media use in communities during periods of stability,
characterized by local responses to routine and periodic emergencies, observe a lack of staff, tools,
policy, and trust restricting social media use among community emergency response and
management officials. In this context, the use contexts and information infrastructures that would
enable the use of social media for community emergency response and management remains
unknown. A gap thus arises between our understanding of the contexts and infrastructures of social
media use during periods of crisis and those that might enable social media use for community
emergency response and management during periods of stability.
Addressing this gap is important for two reasons. First, community officials seek to use
social media for information distribution and monitoring but are prevented by persistent barriers
that result from the lack of staff, tools, policy, and trust among officials in community emergency
management agencies, municipal government, and emergency services. Second, successive
experiences of disaster point to intervening periods of stability as critical time periods during which
communities organize infrastructures- persistent relationships among people and technologies
enabling ongoing and future action- that contribute to resilience during a crisis. By identifying
94
potential alignments between infrastructures enabling social media use for effective emergency
response and management during periods of stability and infrastructures understood to enable
social media use for effective crisis response, this study can outline preparation and mitigation
strategies for building community resilience.
95
Chapter 4
Phase II: Awareness
To effectively collect social media data that can support situational awareness among crisis
responders and affected citizens during a crisis has long motivated researchers and systems
designers (Vieweg et al., 2010). In the case of Twitter, efforts have been made to collect tweets
providing situational reports of events “on the ground” in order to assess damage caused by
earthquakes (Avvenuti et al., 2014), gauge flood levels (de Albuquerque, Herfort, Brenning, &
Zipf, 2015), detect power outages (Bauman, Tuzhilin, & Zaczynski, 2017; LaLone et al., 2017),
and support the work of crisis responders and digital volunteers (Cobb et al., 2014; Hughes, &
Shah, 2016).
However, existing methods to collect situational reports provide only a partial view of all
crisis-related information posted on social media. In the case of Twitter, typical data collection
methods rely on sparse geographic metadata and crisis-related keywords that return a fraction of
all potentially-relevant tweets (Morstatter, Lubold, Pon-Barry, Pfeffer, & Liu, 2014; Saleem et al.,
2014; Shelton et al., 2014). Consequently, “data sets must get bigger… before they can be sampled
or filtered accordingly,” Palen and Anderson (Leysia Palen & Anderson, 2016) explain, “the
bounds of observation must be done through decisions—which may have acknowledged
limitations—to scope the data.” To widen observation of disruptive events occurring on the ground,
crisis responders require new methods to collect more data than now available and, at the same
time, better understanding of the limitations of each method so that multiple methods can be
combined in ways that expand awareness during a crisis.
This study contributes to the critical examination of big crisis data (Carley et al., 2016;
Mulder, Ferguson, Groenewegen, Boersma, & Wolbers, 2016; Leysia Palen & Anderson, 2016) by
96
comparing existing location and keyword filtering methods with a new data collection method-
network filtering- to show how each conditions particular opportunities for situational awareness
during a hyperlocal severe weather emergency. Our findings offer two primary contributions.
First, this study introduces Social Triangulation, a novel data collection method that uses
social network ties to infer Twitter users living in a geographic community and collect tweets they
post during a crisis. Social Triangulation involves a four-phase process to locate social media users
in a geographic area: (1) categorizing community assets, (2) cataloguing community assets, (3)
collecting user information, and (4) analyzing geographic characteristics and information curation
patterns. Employing Social Triangulation to identify local users in State College, Pennsylvania, this
study infers the location of 54k social media users associated with the geographic community.
Table 4-1. Phase II research questions.
How can emergency response officials identify hyperlocal social media users and collect
hyperlocal social media data?
RQ1 What information curation behaviors characterize social media users following
community organizations?
RQ2 What is the relationship between information curation behavior and social media
users’ self-identified geographic location?
RQ3 What is the distribution of situational awareness information across location, keyword,
and network-filtered social media data during an emergency?
RQ4 What is the distribution of unique incidents across location, keyword, and network-
filtered social media data during an emergency?
Second, this study deploys and compares Social Triangulation/network filtering with
existing methods during a hyperlocal weather emergency to find that over half (52%) of all
situational reports are ignored when using only location and keyword-based methods to collect
social media data during a crisis. Findings show that each of the three methods identify unique
97
incidents of infrastructure damage and service disruption reported on Twitter, but network filtering
alone identifies nearly three quarters (73%) of all incidents reported during the emergency. These
findings suggest that combining multiple data collection methods is necessary when using Twitter
to support situational awareness during a crisis. The two studies of Phase II are guided by one
primary and four secondary research questions (Table 4-1).
Identifying Hyperlocal Social Media Users and Information Sources
At the same time that citizens in a geographic community use social media platforms such
as Twitter to access and share critical information during emergencies (Mazer et al., 2015; Olteanu
et al., 2015; Starbird et al., 2010), community public safety officials use social media to distribute
timely information and detect information to support the situational awareness needs of first
responders (Denef et al., 2013; Grace et al., 2018; Hiltz & Plotnick, 2016; Hughes, & Palen, 2012).
However, existing location and keyword filtering methods to collect real-time social media data
during a crisis remain limited by the sparsity of geographic metadata associated with tweets, and
the tendency of keyword-based methods to capture highly-visible information posted by remote
rather than local users (Bruns & Liang, 2012; Carley et al., 2016; Morstatter et al., 2013; Olteanu
et al., 2014). As a result, community emergency responders end up collecting only a small
proportion of hyperlocal data created by a small proportion of citizens in a geographic area.
Moreover, by collecting social media data from only those users who post information, users who
use social media to access rather than share information remain invisible to community emergency
responders distributing alerts, warnings, and situational updates before, during and after an
emergency.
98
The limitations of location and keyword filtering demand new methods for both identifying
hyperlocal social media users and collecting the data they create in order to effectively distribute
official information and identify situational awareness information during an emergency. This
study introduces Social Triangulation and network filtering as novel methods for inferring
community networks of social media users located in a geographic area and collecting hyperlocal
social media data produced by community networks during an emergency, respectively.
Local Social Media Users
Local social media users represent an important source and audience for information during
emergencies. While local users (hereafter simply “locals”) share and access social media content
from locations within a geographic community. they pose a challenge of discovery for community
officials attempting to distribute information to the public and detect information that can support
the situational awareness needs of emergency responders. As Landwehr and Carley (2014) write:
Locals at the site of the disaster who are posting information about what they are witnessing
are in many ways the gold of the social media world, providing new, actionable information
to their followers. They are few in number, and while their messages are sometimes
reposted they often don’t circulate broadly. Locating their content is an ongoing challenge
akin to finding a needle in a haystack. (p. 2)
In the field of Crisis Informatics, Locals using social media platforms such as Twitter have been
previously identified through post hoc content analyses that attempt to locate users using
geographic references included in tweet content (Olteanu et al., 2015; Starbird & Palen, 2012), or
real-time keyword and location filtering using emergency-related search terms and geographic
coordinates (i.e. geotags) that append some tweets (de Albuquerque et al., 2015; Kogan et al.,
2015). While content analyses and location filtering locate users based on select emergency and
geographic references, geotags affix only 1-3% of all tweets (Morstatter et al., 2013). Self-
99
identified location information some users include in their profile presents an alternative, however,
some users do not disclose their location, or decide to include ambiguous or non-geographic
information instead (Wang, Harper, & Hecht, 2014). Moreover, each of these methods ignore locals
accessing but not sharing information- the silent majority emergency responders target when using
social media for crisis communication (Houston et al., 2015; Rice & Spence, 2016).
To identify locals, these methods establish important conditions for localness: social media
users identified as locals must share tweets (often after an emergency has occurred) and use explicit
geotags or keywords (Starbird et al., 2010). Although location and keyword-based data collection
and sub-sampling methods have been advanced (Olteanu et al., 2014), existing methods ignore the
majority of locals who do not share geotagged information or include explicit emergency and
location-related keywords, and who use social media to access rather than share information.
Consequently, these locals remain invisible to community emergency responders seeking to
communicate warnings and gain on-the-ground information before, during, and after an emergency.
As a solution, this study introduces a novel method to identify local social media users
called Social Triangulation (ST). ST is motivated by the basic assumption that people using social
media and living in a geographic community curate their social networks to include local
organizations both physically present in the area and disseminating information on social media
concerning that area. In the case of Twitter, this assumption suggests local people follow local
organizations. The value of this method is to twofold: ST identifies hyperlocal social media users
accessing rather than sharing information, as well as the hyperlocal information sources they access
before emergencies occur. Similar to methods of community asset mapping performed by urban
planners (Kerka, 2003; Mcknight & Kretzmann, 1997), ST involves first categorizing and
cataloguing a comprehensive list of community organizations maintaining social media accounts
to discover users following these organizations, and then evaluates the likelihood that these users
100
are indeed locals. Furthermore, analyzing how locals curate their social networks to receive
different types and varieties of local information reveals the information infrastructure of a
community and can inform effective emergency communications strategies.
This study first describes the method of ST and then evaluates the validity of its basic
assumption—that social media users located in a geographic community follow organizations
within that community. In an exploratory application of ST within a small city in the Northeastern
United States, 195 community assets are catalogued and categorized to identify 185,176 Twitter
followers whose location is then evaluated using users’ self-identified location information. This
method finds promising evidence supporting the basic assumption of ST: among 79,998 users who
have location information on their Twitter profiles, fully 68% identify themselves as
locals. Moreover, the more community organizations a user follows the more likely they are to
identify as locals: for followers of only one organization, 67% identify as locals, while 71% of
those following 3-9, 84% of those following 10-49, and 98% of those following 50 or more
organizations, respectively, identify as locals.
Additionally, this study explores the information infrastructure of the community by
examining the types and variety of community organizations people follow on Twitter to access
local information. This exploratory study points to differences in the level and position in which
citizens are embedded in the information infrastructure of a community. Specifically, a majority of
loosely embedded users tend to curate and receive information from a single local media source,
while a deeply embedded minority is positioned to receive more and various kinds of community
information relative to the latter, especially information disseminated by citizens’ associations,
civic government agencies, and emergency services.
This study therefore finds utility in ST as a method supporting community emergency
communications planning and resilience-building. By systematically cataloguing and mapping
101
community assets that act as important sources of local information on social media, as well as the
local social media users who access information shared by them (as well as the locals who do not),
ST can inform emergency communication planning by pointing to groups of locals not directly
receiving emergency-related information, as well as the social media accounts of “community
influencers” well-positioned to reach these groups by re-distributing official information to their
follower networks. Drawn from this study, three emergent guidelines for emergency
communication with social media are presented: i) account for local filter bubbles, ii) identify
community influencers, and iii) cooperate with citizens’ associations.
This presentation of ST is guided by two exploratory research questions: (RQ1) What
information curation behaviors characterize social media users following community
organizations? (RQ2) What is the relationship between information curation behavior and social
media users’ self-identified geographic location? To answer these questions, this study is organized
as follows: first, work in Crisis Informatics is reviewed to point out that the majority of locals using
social media, especially as information recipients, tend to be excluded by current data collection
and localization methods. Second, the procedure of ST is presented in a case study of the
information infrastructure of a small city in the Northeastern US. The type of data ST offers related
to local information curation behaviors are then discussed, followed by an evaluation of the
localness assumption behind ST. This study concludes by presenting emergent guidelines for
emergency communication planning with social media and discussing opportunities for future
research.
102
Why is identifying locals important?
Local social media users, or locals, inhabit a geographic area and constitute communities
in which they share and access information. In Crisis Informatics research, locals are those affected
by disasters and more routine emergencies and take part as eyewitnesses and citizen responders.
Locals share situational awareness information for other locals and emergency responders, as well
as, in the case of disaster, a global crowd of interested observers, journalists, and digital volunteers
(Starbird, Muzny, & Palen, 2012; Starbird et al., 2010). Moreover, locals live and work in towns,
cities, and regions where they intimately know the geographic, cultural and built infrastructure, and
can therefore, unlike remote users, share unique information during a crisis (Starbird et al., 2012).
When sharing information, locals can provide “generative information,” primary accounts
and interpretations of emergency-related events that Starbird and colleagues (2010) take as the “the
core of the information production cycle” (p. 246). Studies identify locals as individuals who are
more likely post messages related to emergencies in their vicinity than social media users
geographically-removed from events (de Albuquerque et al., 2015; Lachlan, Spence, Lin, Najarian,
& Del Greco, 2016). Moreover, suggesting the importance of social networks for re-distributing
information, studies observe that locals are more likely retweet generative information created by
other locals than geographically-remote users (Kogan et al., 2015; Starbird et al., 2010). Whereas
locals often share generative, unique information, the globally-distributed crowd typically post
messages of sympathy and support for those locally affected by a crisis (Olteanu et al., 2015).
Nevertheless, remote users- by retweeting- can amplify and direct attention information locals
share, and re-distribute relevant information across online communities (Starbird & Palen, 2012).
When accessing information, locals constitute those social media users whom government
and non-government response organizations attempt to communicate situational updates and
directives during crises. Studies find emergency response agencies adopt social media in
103
emergency situations (Graham, Avery, & Park, 2015; Hughes, & Palen, 2009), and seek to use
social media to disseminate warnings to locals before emergencies (Rice & Spence, 2016; Veil et
al., 2011). During and after emergencies, governments and NGOs post social media to convey
important information to affected locals on the progress of relief operations, and the location and
availability of emergency assistance (Olteanu et al., 2015; Tim et al., 2017). In non-emergency
contexts, municipal governments and emergency services, such as police departments, seek to
communicate crime and traffic updates among locals, and solicit information about missing persons
and wanted suspects (Huang, et al., 2016).
Despite the importance of locals to crisis information cycles, in the globally-expansive
information space of social media identifiable local social media users remain rarities (Starbird et
al., 2010). Analyzing social media posted across multiple natural and man-made disasters, Olteanu
et al. (2015) find that information shared by locals account for approximately 9% of all crisis-
related messages on Twitter. Moreover, information locals share often lacks the content features-
hashtags, toponyms, or crisis-related keywords- that make information visible and accessible to
remote users and emergency responders (Bruns & Liang, 2012; Saleem et al., 2014; Vieweg et al.,
2010). As identifying hyperlocal social media users sharing and accessing information remains
“akin to finding a needle in a haystack,” community emergency responders require effective
localization methods to inform emergency communication with social media before, during, and
after a crisis.
How have locals been previously identified?
Typical methods in Crisis Informatics collect social media data via the words locals include
in tweets and/or geolocation information associated with tweets or users’ profiles. These methods
104
typically involve engaging Twitter’s public API and can be referred to as keyword or location-
based collection respectively (Olteanu et al., 2014, p. 376). Importantly, these collection methods
set important conditions for identifying locals.
Location-based Collection and Localization
Locals can be identified based on geographic coordinates (i.e. geotag) associated with a
tweet, or location information included in a user’s profile. Using Twitter’s public API, queries can
be constructed to retrieve tweets containing keywords, hashtags, or tweets with a location-identifier
corresponding to a geographic area or bounding box specified by a set of geographic coordinates.
Due to issues of relevance and comprehensiveness involved with using Twitter’s REST and
Streaming API, keyword-based data collection often precedes location-based subsampling and
subsequent identification of locals (Kogan et al., 2015).
For instance, to identify geographically vulnerable Twitter users during Hurricane Sandy,
Kogan et al. (2015) first selected eight “carefully chosen” keywords to collect tweets through
Twitter’s Streaming API. Among the users whose tweets were collected, users posting at least one
geotagged tweet within a boundary box encompassing part of the United States Eastern Seaboard
were selected as the basis for a subsequent round of data collection involving the REST API, by
which the authors could collect up to 3200 of these users’ most recent tweets (p. 983). This
approach to identifying locals thus involves two important filters: those who use a specified set of
keywords in their tweets and, the small minority among those users who also posted a geotagged
tweet in the affected locale.
Using geotags to identify locals remains limited by the paucity of users who include
geographic coordinates in their tweets- only 1-3% of tweets carry geotags (Morstatter et al., 2013).
105
Consequently, locals posting geotagged tweets represent only a fraction of those actively posting
content on Twitter in a geographic area, and none of the people who use Twitter but do not actively
tweet, or at least do not do so during periods of data collection. Among the latter, a PEW Research
Center (2015) study found that 28% of Twitter users did not tweet during a one-month period, and
an additional 33% posted fewer than nine messages. Among those who did tweet, 30% of all tweets
were retweets (for news-related tweets this rose to 49%) - a notable figure when considering that
location-based queries to the Streaming API do not retrieve retweets (Twitter, 2016).
Content and Keyword-based Collection and Localization
Alternatively, content and keyword-based methods revolve around the relationship
between the lexical scope and distribution of words curated by both locals and a global crowd
during a crisis, and the selection of specific keywords held to represent and, in effect, constitute the
crisis information space. The users whose tweets contain these keywords and hashtags can
subsequently be identified as locals by using tweet geotags or profile location information, as
described above, or coding methods to determine a user’s location based on the content of the
tweets they post, for instance, if it is judged to be an “eyewitness” account (Olteanu et al., 2015).
Using the Streaming API to examine the 2011 Egyptian political protests, for example,
Starbird and Palen (2012) relied on Arabic speakers in their research lab to select certain keywords
“as the most popular during the early days of this event” (p. 9). Thus, if a local protester, even if at
the center of Tahrir Square, did not include one of the keywords “egypt,” “#egypt,” or “#jan25,”
they would never be identified by the authors and, more significantly, perhaps other locals and the
global crowd. Starbird and Palen (2012) coded the retweets of active users on the basis of
“assertions of or references to being in Cairo during the period of the protests.” Accordingly, users
106
were assessed as either in Cairo, in Egypt but not in Cairo, or outside Egypt. The authors identified
locals as any Twitter user posting a tweet including at least one of the three queried keywords and
judged as being in Cairo at least once within the defined collection period (p. 10).
Keyword-based methods necessarily focus on highly-visible terms that ignore tweets
missing the keywords queried and the users who post them or do not tweet at all (Bruns & Liang,
2012; Olteanu et al., 2014; Vieweg et al., 2010). Keyword curation can amplify the non-local, as it
“assume[s] to establish a dataset of the most visible tweets relating to the event in question, since
it is the purpose of topical hashtags to aid the visibility and discoverability of Twitter messages”
(Bruns & Liang, 2012). While keywords and hashtags stand to benefit those affected in the vicinity
of a crisis, the emphasis on visibility and discoverability especially aids the geographically-
removed crowd who generally lack existing social network connections to local people and
organizations.
Moreover, Vieweg et al. (2010) suggest that familiarity with a locale, such as possessed by
locals, can lead users to omit the very keywords or hashtags that would identify them to others as
local. What they describe as “markedness” refers to:
how certain places, landmarks or items become taken-for-granted and expected when
referred to in more general terms. The RR data set was collected based on search terms
“red river” and “redriver”, and within this data set, if someone mentioned “the river” or
“the flood level” it was commonly understood to be about the Red River, which makes the
Red River “unmarked”— no detail is necessary when referring to it. (p. 1086)
Thus, if someone only provided information on “flood level” such data would only become visible
to the authors if, in another tweet, they had also mentioned one of the two keywords or hashtags
queried for data collection. During their retrospective examinations of three distinct crises events,
a school shooting, tornado, and flood, Saleem, Xu, and Ruths (2014) find that initial tweets
concerning these events often fail to include references to places or the type of emergency, as well
as visible hashtags associated with the event. The authors conclude that “the first tweets carrying
107
situational information tended to lack the kind of identifying keywords and hashtags that would
make them easy to discover in a full Twitter stream” (p. 156).
Taken together, location and content/keyword-based methods set important conditions for
localness. These methods risk omitting the 97-99% of users without geotags (Leetaru, Wang,
Padmanabhan, & Shook, 2013; Morstatter et al., 2013), those sharing information without including
identifiable keywords, and those who do not actively share information but use social media to
access local information.
Geolocation Inference Methods
Prior research also explores geolocation inference methods that can predict the geographic
location of a social media user based on their social network relationships (Backstrom, Sun, &
Marlow, 2010; Jurgens, Finethy, Mccorriston, Xu, & Ruths, 2015; Zheng, Han, & Sun, 2017). In
the case of Twitter, for example, geolocation inference methods generally attempt to predict users’
locations based on a known location established for other users within their social network.
Geolocation inference methods therefore all require “(1) a definition of what constitutes a
relationship in Twitter to create the social network, and (2) a source of ground truth location data
to use in inference” (Jurgens et al., 2015). For the former, prior methods define this relationship as
following, reciprocal following (Davis Jr., Pappa, Oliveira, & Arcanjo, 2011; Li, Wang, Deng,
Wang, & Chang, 2012) and reciprocal mention relationships between users (Compton, Jurgens, &
Allen, 2014; Jurgens, 2013) that can be further characterized by such measures as friends held in
common (Kong, Liu, & Huang, 2014) and social influence (i.e. number of friends’ followers)
(McGee, Caverlee, & Cheng, 2013).
108
For the latter, the set of users whose location is already known is typically established by
using one or a combination of self-reported user location data (i.e. user profile information) (Li et
al., 2012; McGee et al., 2013) and/or geographic coordinates or location-associated keywords
included in tweets posted by users (Cheng, Caverlee, & Lee, 2010; Davis Jr. et al., 2011; Zheng et
al., 2017). These users and their location data then serve as the premise for inferring the locations
of other users who lack this data. Here the same scarcity and ambiguity of geographic data imposed
by Twitter (Hecht, Hong, Suh, & Chi, 2011; Morstatter et al., 2013) provides the preconditions
(and raison d'être) for geolocation inferencing of users. In a meta-analysis of geographic inference
methods, Jurgens et al. (2015) find parameters related to these limitations on ground truth data to
significantly impact predictive accuracy and the proportion of users for whom the location can be
inferred.
Prior approaches turn to these sparse sources of ground truth data because they are the only
sources available for automated extraction through the Twitter API and because they do not have
external information sources to establish the locations of users. However, ST differs from prior
approaches by manually ascertaining ground truth data by drawing on community-based
information resources to catalogue the local Twitter accounts of community organizations (i.e.
ground truth) based on which people using Twitter for community information can be inferred (i.e.
following relationships between people and community organizations).
ST therefore constitutes a specific variety of geolocation inference method concerned with
constructing a ground truth data source without relying on geotag and profile location data, or
toponyms pulled from tweet content. This study introduces ST as a form of community work
supporting emergency communications planning by identifying community information resources
and understanding local people who access them. Compared with existing approaches in crisis
informatics, ST enables a community to identify local social media users i) before emergencies
109
occur, ii) who share and access local information), and iii) in greater numbers than when relying
on keywords and geographic metadata.
Research Questions
This study introduces and evaluates the method of Social Triangulation (ST). In contrast
to existing location and keyword-based methods ST uses community networks created when
Twitter users “follow” the Twitter accounts of community organizations. In contrast to prior geo-
inferencing methods, ST uses the accounts of community organizations to construct a ground truth
dataset in relation to which hyperlocal social media users can be inferred (thus avoiding the need
to rely on sparse geographic metadata that already constrains location-based filtering). As an
exploratory study, the premises for ST must be first examined and evaluated. First, to understand
the number and variety of community organizations social media users curate within their social
networks, this study asks: (RQ1) What information curation behaviors characterize social media
users following community organizations? Second, to evaluate the geo-inferencing assumption of
ST, this study asks: (RQ2) What is the relationship between information curation behavior and
social media users’ self-identified geographic location?
Collecting Hyperlocal Social Media Data
Collecting real-time Twitter data during a crisis typically involves two primary methods
(Olteanu et al., 2014). The first, location filtering, uses Twitter’s Streaming API to return a sample
of tweets (≤1-3% all tweets worldwide) including geographic metadata, latitude/longitude
coordinates associated with a GPS-enabled device (e.g. smartphone) or user-tagged “place,” that
110
fall within a geographic bounding box. The second, keyword filtering, uses Twitter’s Streaming
API to return any tweets that include selected crisis and place-related keywords (including
hashtags). Based on the affordances of Twitter’s Streaming API, these two methods have become
de facto standards for collecting social media data, however, other methods are possible. A third
and hitherto untried method, network filtering, infers the location of users via social network ties
associated with a geographic area to collect tweets from networks of these users located near a
crisis. Importantly, each method introduces limitations for data collection that, in turn, shape
opportunities for situational awareness during crises.
Bias of Geographic Metadata
To collect information from people in crisis-affected areas, crisis informatics researchers
often first filter tweets by location, and then apply subsequent filters to identify situational reports
(de Albuquerque et al., 2015). However, location filtering identifies only tweets including
geographic metadata, a mere fraction- 1-3%- of all tweets posted (Morstatter et al., 2013). Location
filtering thus excludes up to 97-99% of tweets posted during a crisis.
Moreover, studies show that geotagged tweets provide a biased representation of Twitter
user activity (Carley et al., 2016; Hecht & Stephens, 2014; Malik, Lamba, Nakos, & Pfeffer, 2015),
to include the types of information users post in a geographic area (Shelton et al., 2014). Per capita,
more users post geotagged tweets in cities than rural areas, and tend to be younger than the general
population (Hecht & Stephens, 2014; J. Lin & Cromley, 2015; Malik et al., 2015). Uneven tweeting
activity during a crisis can, in turn, bias representations of events occurring on the ground (Carley
et al., 2016; Mulder et al., 2016; Shelton et al., 2014). Separate studies of Twitter activity in and
around New York City during Hurricane Sandy, for instance, observe increased geotagged tweeting
in urban centers damaged by the storm, but relatively sparse Twitter activity in neighboring urban
111
areas that were, in some cases, more adversely affected (Hecht & Stephens, 2014; Shelton et al.,
2014). Shelton et al. (Shelton et al., 2014) conclude “that places on the spatial periphery of the
metropolitan area, e.g., Staten Island or the Bronx, are more likely to be marginalized within data
shadows than more central locations, e.g., Manhattan and Brooklyn” (p. 173).
Using linguistic features to identify non-geotagged tweets posted within the New York
City metropolitan area, Hecht and Stephens (Hecht & Stephens, 2014) discover reports of flooding
in the neighboring city of Hoboken, New Jersey that are missing from geotagged tweets posted in
that area. Among geotagged tweets posted in Hoboken, however, the authors find reports of
flooding in Manhattan (e.g. flooding of New York Times building). By revealing the sparsity of
geotagged tweets and a reporting bias favoring incidents in urban centers over peripheral locations,
these studies suggest that location filtering alone likely fails to identify the breadth and local
diversity of situational reports posted on Twitter.
Bias of Keyword Filtering
Researchers also commonly employ keyword filtering methods to gather tweets by
constructing queries that seek to match select crisis and place-related keywords with words people
are likely to include in tweets during a crisis (Olteanu et al., 2014; Leysia Palen & Anderson, 2016).
Consequently, keywords must be selected that are common among crisis-related tweets and
relatively unique compared to all tweets posted globally to comprehensively gather relevant data
while preventing rate limiting and levels of noise that can quickly become prohibitive when
filtering the global Twitter stream.
The necessary balance between recall and precision, however, often introduces bias
towards course-grained geographic information (e.g. keywords matching city rather than street
names) and information posted by geographically-remote users or oriented to them (Vieweg et al.,
112
2010). For this reason, using combinations of crisis-related words, hashtags, and globally-distinct
place names (Olteanu et al., 2014; Vieweg et al., 2010), keyword filtering collects “the most visible
tweets relating to the event in question, since it is the purpose of topical hashtags to aid the visibility
and discoverability of Twitter messages” (Bruns & Liang, 2012). As a result, the use of keywords
aids the discovery of information about a crisis, but often that posted and consumed by remote
crowds lacking direct ties with people located in crisis-affected areas (Kogan et al., 2015).
Examining multiple keyword-filtered crisis datasets, Olteanu et al. find that eyewitness reports
account for approximately 9% of all crisis-related tweets (Olteanu et al., 2015).
Conversely, Vieweg et al. (2010) observe that tweets posted in crisis-affected areas often
lack visible keywords associated with the event as people living in a geographic area often assume
a shared context:
…certain places, landmarks or items become taken-for-granted and expected when referred to
in more general terms. The... [dataset] was collected based on search terms “red river” and
“redriver”, and within this data set, if someone mentioned “the river” or “the flood level” it was
commonly understood to be about the Red River, which makes the Red River “unmarked”—
no detail is necessary when referring to it. (p. 1086)
Tweets about the “flood level,” for instance, would never be collected unless a user also included
at least one of the two selected keywords. As a result, keyword filtering often excludes situational
reports that lack the globally-visible, course-grained toponyms that tend to be assumed among
Twitter users in a geographic area. Analyzing Twitter activity across three crisis events- a tornado,
flood, and school shooting- Saleem, Xu, and Ruths (Hecht & Stephens, 2014) find that “the first
tweets carrying situational information tended to lack the kind of identifying keywords and
hashtags that would make them easy to discover in a full Twitter stream.”
113
Deploying Network Filtering
After comparing location and keyword filtering methods, Carley et al. conclude that “they
miss most of the user population, and hence may miss critical information about who needs what
help. Improved procedures for inferring location based on the user ties... are needed” (Carley et al.,
2016). Despite established research on geolocation inferencing (Hecht & Stephens, 2014; Jurgens
et al., 2015; Zheng et al., 2017), methods that use social network ties to infer the location of social
media users, crisis informatics research has not adopted this approach to collect data from networks
of Twitter users inferred near a crisis. We refer to this third method as network filtering.
Applied to Twitter, geolocation inference methods have been used to predict a users’ home
city (associated with a geographic area) by comparing social network relationships among users
whose locations are known (e.g. users who post geotagged tweets) and unknown (Jurgens et al.,
2015; Zheng et al., 2017). Someone who follows Twitter accounts followed by many people known
to be living in the same geographic area, for example, may be inferred to also live in that area
(McGee et al., 2013).
The limitations of location and keyword filtering recommend new methods of data
collection that can capture some of the 97-99% of tweets lacking geotags, as well as compensate
for the urban and global biases associated with each method, respectively. In an approach we refer
to as network filtering, geolocation inferencing methods can be adopted to identify and collect
social media data from networks of users associated with a geographic area. Unlike location and
keyword filtering, network filtering relies on neither geographic metadata or the content of tweets
to geolocate information posted on Twitter and might be deployed to collect more and more diverse
geolocated Twitter data than now possible using location and keyword-based methods. However,
lacking empirical evidence, the relative utility of network filtering in this respect remains unknown.
114
Research Questions
In this follow-on study we deploy and compare all three data collection methods- location,
keyword, and network filtering- to analyze the opportunities they afford when constructing
situational awareness during a crisis. In the context of a severe storm in Centre County, United
States, we consider the following questions: (RQ3) What is the distribution of situational
awareness information across location, keyword, and network-filtered social media data during an
emergency? (RQ4) What is the distribution of unique incidents reported in situational awareness
information across location, keyword, and network-filtered social media data during an
emergency?
Method: Social Triangulation and Network Filtering
This section describes, first, the methods employed for Social Triangulation, including
community asset mapping of community organizations’ Twitter accounts, social network analysis
measures (e.g. E-I index scores) to examine the information curation behaviors of Twitter users’
following community organizations, and use of Google Fusion Tables to evaluate users’ self-
identified location information. Next, this section describes the employment of location, keyword,
and network filtering to collect Twitter data during a severe weather event on May 1st, 2017, and
the qualitative content analysis of Twitter data collected during the storm.
Social Triangulation
ST is here deployed by cataloguing community assets located in a small city in the
Northeastern US to analyze the information curation behaviors of users who curate Twitter
115
accounts of these community assets within their social networks. Information curation “involves
future oriented activities,” consisting of a “set of practices that select, maintain, and manage
information in ways that are intended to promote future consumption of that information”
(Whittaker, 2011, p. 7). Here, curation involves practices of “following” community assets by
Twitter users who, as a result, enable ongoing access to local, community information disseminated
by these organizations. Shared patterns of curation among people following local organizations
reveal the information infrastructure of a community and reveal groups of local citizens who curate
their social networks to directly access types of local information, as well as those who do not.
Table 4-2. Categories of community organizations (*unique followers).
Category Description # Followers
Bars Establishments of good cheer, e.g. bars, saloons, taverns, wineries, and pubs
27 25439
Citizens’ Associations
Volunteer, social, and nonprofit organizations, e.g. habitat for humanity, environmental conservation groups, historical society
41 24778
Civic Services Civic government and public services, e.g. municipal government, public library, public transit
18 26372
Emergency Services
Emergency management and response services, e.g. police and fire departments, EMS, university emergency alerts
13 16833
Entertainment Recreational businesses, e.g. minor league baseball team, golf courses, movie theaters
5 20909
Media Local media, e.g. newspapers, newsletters, news websites, radio, television
32 202182
Restaurants Local (non-chain) food-serving establishments, e.g. cafes, diners, delis, pizzerias
43 15041
Schools Municipal school district, e.g. elementary, middle, and high schools, local vocational school
16 5573
Total 195 185176*
Social triangulation (ST) involves a four-phase process to locate Twitter users in a
geographic area: (1) categorizing community assets, (2) cataloguing community assets, (3)
collecting user information, and (4) analyzing geographic characteristics and information curation
116
patterns. To socially triangulate Twitter users, the authors began by identifying categories of public
and private organizations to organize the search for community assets in one city in the
Northeastern Unites States. Urban planners often assess community capacity and resources using
community asset mapping methods which search for organizations in a community by type (Kerka,
2003; Mcknight & Kretzmann, 1997). Similarly, the authors identified types or categories of
organizations- public institutions, citizens’ associations, local economy (e.g. businesses), and
media- which were subsequently modified and expanded by searching for and cataloguing
organizations maintaining a Twitter account in the city of interest. The resulting eight categories of
community assets are listed in Table 4-2.
The initial categories functioned primarily to organize the search for local organizations.
This involved a grounded process in which online directories and search engines were used to
explore the diversity of local organizations in a community asset category and determine which
types of organizations in that category maintain an identifiable social media presence. For example,
beginning with the community asset category of local economy the authors used community
directories (e.g. yellow pages) and general online searches (e.g. “restaurants in...”) to categorize
and identify types of businesses maintaining Twitter accounts. The authors subsequently identified
three categories- restaurants, bars, and entertainment. Importantly, the categories reflect local
economic assets with a social media presence and not a comprehensive catalogue of community
businesses. For example, beginning within the community asset category of citizens’ associations,
the authors initially speculated that the numerous churches and worship centers in the city might
have a significant social media presence, however, upon investigation, very few were on Twitter.
The second stage of ST involved developing a comprehensive catalogue of organizations
for each category. The grounded process of identifying local organizations and arranging them into
categories described above yielded an extensive catalogue of organizations, however, the authors
117
sought to develop, to the best extent possible, a listing of all identifiable organizations both located
in the city and maintaining a Twitter account for each category selected. Additionally, a project
Twitter account was created to utilize Twitter recommendation algorithms to suggest additional
organizations and searched among the followers of these organizations, as well as the accounts
each organization follows, to identify any organizations that did not appear in online searches.
Altogether 195 organizations were catalogued across eight community asset categories with Twitter
followings ranging from six people following a small pizzeria to the 110,000+ followers of an
online news site (Figure 4-1).
Figure 4-1. Twitter users following community organizations by asset category.
Third, using Twitter’s REST API the authors collected the Twitter IDs of accounts
following each local organization. For example, among civic services the city’s bus system was
catalogued and the account IDs for the 1,011 people that follow the official bus system account on
Twitter were collected. In total, 185,176 Twitter IDs were collected, each following at least one of
118
the 195 local organizations catalogued. The follower counts for each organization reflect the period
of data collection, which occurred during the week of December 5, 2016.
The fourth phase of ST involves evaluating the localization assumption behind ST and
analyzing the kinds and variety of local information users access through their social networks. For
the latter this study analyzes, 1) the curation patterns among those following community
organizations and 2) the relationships among organizations with respect to users’ curations patterns.
The two analytical methods are further described below.
Geographic Analysis
Profile locations of the 185,176 Twitter users were utilized to understand their self-
identified location relative to the number of organizations they follow. A Java program written to
search the Twitter API found Twitter accounts available for 168,452 of these accounts. Of these,
79,980 accounts included profile information. By importing this information into Google Fusion
Tables, the authors geocoded each of the self-described locations by relying on similar tools that
are used to identify non-uniform information in Google Map search. Errors in geocoding can occur
due to misspellings in the location, and ambiguous locations which match multiple, similarly named
places in the world (Wang et al., 2014). Across multiple sessions of Google’s geocoding process,
between 12 – 15% of the locations were reported as ambiguous and unavailable for geocoding. The
identifiable locations were plotted on a Google Map to better understand the locations self-
identified by Twitter users with respect to the number of local organizations they follow.
119
Social Network Analysis
Using the data collected, a 2-mode matrix was constructed where one mode was comprised
of the Twitter accounts of each of the 195 local organizations and the second mode was comprised
of the 185,176 Twitter users that followed those organizations. With an interest in uncovering
patterns in information curation behaviors among various data streams, a weighted 1-mode
affiliation matrix was created that represents Twitter user co-followership of the 195 local
organizations. In this network, each node is a local organization and each link among organizations
is weighted by the number of Twitter users that co-follow both organizations.
Organizational categories were used as node attribute information in order to describe the
relationships among types of local communities. Further, a method to examine an attribute-based
model of structuration of the local network was developed. This was performed by first creating a
table of homophily scores using a standard E-I index. Here, each “same category” tie is treated as
an internal group tie and every “different category” tie is treated as an external group tie, in which
the number of ties external to the groups (E) minus the number of ties that are internal to the group
(I) are divided by the total number of ties in the network (Krackhardt & Stern, 1988). The
homophily scores table was then treated as a 2-mode matrix to create a 1-mode affiliation matrix
representing inter-category ties among various types of Twitter information streams resulting in a
community asset network.
Location, Keyword, and Network Filtering
This section describes the employment of location, keyword, and network filtering to
collect Twitter data during a severe weather event on May 1st, 2017, and the qualitative content
analysis of Twitter data collected during the storm.
120
Location and Keyword Filtering
Location filtering involved the use of Twitter’s Streaming API to collect tweets within a
bounding box encompassing Centre County, Pennsylvania during a twelve-hour period (12pm-
12am) before, during, and after a severe storm and tornado that struck the area on May 1st, 2017.
This produced the Location Dataset totaling 17,849 original tweets including either
latitude/longitude coordinates (i.e. geotag) or user-tagged places located within the county.
Keyword filtering was also performed to filter tweets that include 48 place names,
including “Centre County” and the names of its 47 municipalities, boroughs, and census-designated
places.1 Data was collected for the 12-hour period to produce the Keyword Dataset totaling 9455
tweets.
Network Filtering
To infer and collect tweets from networks of users in Centre County we deployed a simple
geolocation inferencing method that we introduce as a novel network filtering technique to collect
Twitter data posted within a geographic community (Grace et al., 2017). Typical geolocation
inference methods attempt to infer n-locations for a set of Twitter users, and require “(1) a definition
of what constitutes a relationship in Twitter to create the social network, and (2) a source of ground
truth location data to use in inference” (Jurgens et al., 2015). Most approaches utilize following or
mention ties among Twitter users (Hecht & Stephens, 2014; Zheng et al., 2017) and geographic
metadata, geographic references in tweet content, or profile location information as the source of
ground truth for inferring the locations of users lacking geographic information (Hecht & Stephens,
1 http://centrecountypa.gov
121
2014; McGee et al., 2013). Lacking external information sources to seed the network with ground
truth user locations, these approaches rely on these sparse sources of ground truth data (e.g.
geotagged tweets) because they are the only sources available and suitable for automated extraction
using the Twitter API.
As we seek to infer n-users for only a single geographic location (e.g. Centre County) we
can approach the geolocation inferencing problem differently. Exploiting the tendency of local
people to follow local organizations (McGee et al., 2013), we ascertained ground truth data by
manually cataloging 195 Twitter accounts belonging to categories of organizations located in the
county (bars, civic and emergency services, citizens’ associations, entertainment, media, schools,
restaurants) in order to identify and extract their networks of followers. These procedures are
described in detail in (Grace et al., 2017). We extracted account IDs for 185,176 users and, as our
approach initially prioritizes recall over precision, we began continually collecting all tweets posted
by this network of users via Twitter’s Streaming API beginning in March 2016. Using this network
filtering technique, data was collected during the 12-hour period of the storm on May 1st, producing
the Network Dataset totaling 17351 tweets.
We evaluated the accuracy of this broad but potentially coarse-grained inferencing
approach in two ways. First, to evaluate if most users were in the area of interest, we used Google
Fusion Tables to geocode and compare the profile locations of approximately 80k users who self-
entered an identifiable location on their profiles with our geographic area of interest. Among these
users, 68% entered a profile location within the county, and over 90% within the state. These results
indicated that the network of users significantly overlaps with users located in the county, and that
tweets posted by this network would be likely to provide situational information during the storm.
Second, during our qualitative coding process, we manually investigated every tweet providing a
122
situational report from all three datasets to determine if the post provided local information. We
discuss this process in detail below.
Qualitative Content Analysis
We manually coded each tweet of the Location, Keyword, and Network datasets to
understand the types of information posted on Twitter during the 12-hour period of the storm and
identify tweets- situational reports- that might support situational awareness during the emergency.
Qualitative content analysis provides a grounded and systematic approach for understanding the
diversity of information people report on social media during a crisis, including those that support
situational awareness (Hsieh & Shannon, 2005; Vieweg et al., 2010). This analysis involved three
stages analyzing, in turn, tweet relevance, situational information, and location information.
First, we coded tweets as “on-topic” if any part of the tweet content referred to weather or
its consequences (e.g. damage caused by high winds), and “off-topic” if the content of the tweet
did not. In this initial coding process, we attempted to distinguish between emergency-related, on-
topic tweets and the diversity of off-topic posts that accompany disruptive events (Olteanu et al.,
2015; Vieweg et al., 2010). To ensure coding accuracy a random set of 1000 tweets were first given
to all three coders and a Cronbach alpha test was run yielding α = 0.92. Coding differences were
deliberated and reconciled, and then the entire dataset of 44655 tweets was then subdivided and
coded for relevance, resulting in 3113 (7%) on-topic tweets and 41542 (93%) off-topic tweets.
Second, on-topic tweets were coded for a second time to understand the types of
information reported. Together, the authors engaged in a grounded, iterative process of open coding
that involved assigning meanings, in the form of emergent code categories, to all on-topic tweets
in all three datasets (Hsieh & Shannon, 2005). As an iterative process, we refined our code
123
categories through a process of constant comparison by re-analyzing assigned codes when new
themes emerged throughout the coding process (Glaser & Strauss, 1967). This process involved
the grouping and refinement of categories and sub-categories created during open coding (e.g. axial
coding) (A. Strauss & Corbin, 1990). We eventually arrived at 19 code categories accounting for
the diversity of all on-topic information.
During this process we consulted categories developed in prior content analyses of crisis-
related social media (Olteanu et al., 2015; Starbird et al., 2010). While coding we noticed a diversity
of information reporting forms of infrastructure damage prior studies suggest can support
situational awareness during a crisis, including tweets reporting damage to buildings (Olteanu et
al., 2015), roadways (Hong et al., 2017; Tien et al., 2016), and electrical infrastructure (Bauman et
al., 2017; LaLone et al., 2017). While this work informed our grounded analysis, the data we
encountered revealed types of information that unpacked categories developed in prior research.
For what Olteanu et al. categorize as “Infrastructure & Utilities” (Olteanu et al., 2015), for example,
we develop five distinct categories: property, road, and power line damage, Internet outage, and
power outage. Given the potential utility of this situational information (Bauman et al., 2017), we
focused subsequent analysis on these six categories (Table 4-3).
Third, we assessed if the tweets describing infrastructure damage and service disruption
provided local information, here understood as a description of a physical event occurring in the
geographic area of interest. We recorded these events to establish a catalogue of incidents reported
by Twitter users across the three datasets. To do so, we adapted criteria for determining local
information utilized in prior studies (Olteanu et al., 2015; Starbird & Palen, 2012). We determined
a tweet provided local information if it a) made a geographic reference to a place(s) within Centre
County, was posted by a user who b) posted a geotagged tweet(s) within the county on May 1st, or
c) self-entered a profile location and, in their extant tweet stream, made a geographic reference(s)
124
within the county. During this process we encountered many tweets we determined to be non-local
although they provided information about locations in nearby, adjacent counties. The tweets
ultimately assessed to provide local situational information constitute the Situational Report
Dataset, totaling 352 tweets reporting 44 incidents across the county.
Table 4-3. Coding categories for infrastructure damage and service disruption.
Code Category Description | Example
Power Line Damage Tweets reporting downed or damaged power lines
“our neighbors reported line dwn across [road] & [road] and that was 4 hrs
ago…”
Property Damage Tweets reporting damage to building and property
“storm passes. no problems for us but two neighbors had trees hit their homes”
Road Damage Tweets reporting damage to and obstruction of roadways
“Tree down across [road] near the Meridian. Police have it blocked”
Storm Damage Tweets reporting unspecified damage caused by high winds and rain
“Major tree damage and flooding around the county. Please drive carefully!”
Internet Outage Tweets reporting loss of internet connectivity
“Either the storm is knocking out wifi in [building name] or this place is haunted”
Electricity Outage Tweets reporting the loss or interruption of electricity
“Lights out workout at East Coast Health & Fitness in [place]. literally! #blackout”
For emergency responders using social media to detect situational reports of emergency in
a geographic area requires new methods for collecting hyperlocal social media data. In the case of
Twitter, emergency responders are limited to the 1-3% of tweets with geographic metadata (i.e.
geotags) and tweets containing keywords (e.g. “flood”) selected to filter the global Twitter stream.
While location filtering excludes most tweets posted to Twitter, keyword filtering excludes any
tweet lacking the narrow set of geographic (e.g. city names) or emergency-related keywords (e.g.
flood) that can make information visible during a global Twitter search. In contrast, Social
Triangulation and network filtering supplement these methods by identifying hyperlocal social
media users and collecting hyperlocal social media data, respectively.
125
Employing Social Triangulation to identifying hyperlocal social media users and
information sources in the town of State College, Pennsylvania, this study finds support for the
geo-inferencing assumption that local social media users tend to follow local organizations on
Twitter. Second, comparing social media data collected using location, keyword, and network
filtering during a severe weather emergency, network filtering enables emergency responders to
identify twice as many situational reports of infrastructure damage and service disruption as
location and keyword filtering combined.
Findings: Identifying Hyperlocal Social Media Users and Information Sources
The following section describes the results of the descriptive and comparative analysis of
Twitter users following community organizations identified using ST. A descriptive account of
information gathered using the method is presented, followed with analyses that examine the
geographic location and information curation behaviors of users following community
organizations.
Community Organization Following among Users
ST identifies local social media users who curate community organizations within their
social networks and, as a result, gain access to information shared by those organizations. Among
users curating at least one of the 195 citizen’s associations, public services, businesses, and media
organizations in their social networks, how many and what categories of organizations do they
follow? Figure 4-2 displays the following distribution of the 185,176 users, indicating different
levels of embeddedness within the local information infrastructure: a minority of users follow many
126
community organizations while the majority follow few. On average each user follows 1.82
organizations, with 75.3% (139,440) of all user following only one of the 195 organizations
catalogued in the city.
Grouping the 185,176 users according to how many organizations they follow reveals five
follower levels categorized as: unique, low, moderate, high, and extreme. As described, the
majority users follow only one community organization, users henceforth refered to as unique
recipients of local information. Low followers, 12.4% (23,027) of all users, follow only two
organizations. Medium, 10.5% (19,397), and high followers, 1.7% (3,147), follow between 3-9 and
10-49 organizations respectively. Accounting for only 0.1% (165) of all users, extreme followers
curate 50+ organizations within their social network, such as one user following 139 organizations
in the community. This following distribution indicates, on one hand, a highly embedded minority
of users (medium-extreme followers) accessing multiple information streams within the
community and, on the other, a weakly embedded majority (unique-low followers) that directly
accesing information from only one or two organizations.
Figure 4-2. Number of community organizations followed among Twitter users (colored by
follower level).
127
Looking to the categories of organizations users follow points to important differences in
the kinds of organizations users choose to curate within their social networks. Comparing unique
followers to high and extreme followers, for instance, reveals a stark contrast in the curation of
media organizations among users (Figure 4-3). Among the 139,440 users who follow one
organization, over 106,000- 58% of all users- follow one of 32 local media organizations. In
contrast, medium, high, and extreme followers curate more and more diverse organizations within
their social networks. Among high and extreme followers, citizens’ associations and public services
account for approximately 40% and 50%, respectively, of the organizations curated within their
social networks. That users occupy different positions of embeddedness within the community
information infrastructure indicates that different segments of the community access different
information shared by different sources of community information. Thus a minority of highly
embedded users follow more organizations and access a more balanced variety of information from
media, businesses public services, and volunteer groups. On the other hand, a majority of weakly
embedded users follow few organizations and of less variety, with most accessing information only
from local newspapers, radio stations, or telelvision sources.
Figure 4-3. Categories of community organizations followed among Twitter users (by follower
level).
128
Evaluation of Users’ Locations
To evaluate the localness assumption underlying ST calls for evaluating the relationship
between information curation behavior and users’ self-identified location. That locals will tend to
follow and access information from organizations in their geographic area motivates ST as a
geolocation inference method. To evaluate this assumption, the self-identified location some users
include in their Twitter profile is used to evaluate the location of users following community
organizations and understand how users’ locations vary according to the number of community
organizations they follow (Table 4-4). Of the 185,176 users following a community organization,
43% or 79,978 include an identifiable location in their profile. Among those including identifiable
location information in their profile, 91% are local to the state, and 68% self-identify as local to the
municipality. Overall, 54,165 followers, or 29% of users following at least one community
organization, reside in the city.
Table 4-4. Self-identified location for users following community organizations.
Following Level Total Users Users Located
% In State % In City (25km) %
Extreme (50+) 165 146 88% 146 100% 143 98%
High (10-49) 3147 2142 68% 2058 96% 1809 84%
Moderate (3-9) 19397 10404 54% 9755 94% 7361 71%
Low (2) 23027 10530 46% 9796 93% 7064 67%
Unique (1) 139440 56756 41% 50883 90% 37788 67%
Total 185176 79978 43% 72638 91% 54165 68%
Among all users 72638 39% 54165 29%
129
Differences, however, characterize the relationship between information curation behavior
and users’ self-identified locations. Depending on how many community organizations a user
follows, differences appear in the availability of location information and the proportion of
followers who identify as living in the community. First, among unique followers only 41% include
an identifiable location in their Twitter profile. In comparison, 68% of high followers and fully
88% of all extreme followers include profile locations. Significantly, in the locations identified
among these users, nearly all followers who include location information on their profile are local
to the state (91%), and most are local to the municipality (68%). Moreover, among high and
extreme followers who include profile locations, 84% and 98%, respectively, identify as locals.
Figure 4-4. Geographic locations of users following community organizations.
130
The self-identified location information users provided on their profiles was mapped using
Google Fusion Tables and Google Maps (Figure 4-4). Separately mapping the self-identified
locations of users at different follower levels, dramatic differences in the geographic dispersion of
users’ locations are revealed. Among unique and low followers, users identify locations spanning
around the globe. For instance, a user following a local bar indicates that he is from Iraq, while a
baseball-related twitter account in Cuba follows the city’s minor league baseball team.
Analysis of Local Information Curation Behaviors
Figure 4-5 displays the co-following network structure of the 195 community organizations
that share a link if followed by the same Twitter user. The network is very dense, with 96.6% of all
possible ties present, but has a low overall degree centralization with only 3% of the network
centralized around a few nodes. The network is very tightly bound with a network diameter of two
and the average distance between any two nodes is 1.034. While many whole network measures of
the network are not descriptive, a weighted measure of centrality shows that the whole network
homophily E-I index among organizations linked by the same organizational type is 0.6892. An
exploration of the one-mode representation of the community information infrastructure divided
by follower types finds similarly high densities with low diameter. These measures indicate that
the community appears to have very little fragmentation in its information infrastructure.
The next phase of analysis utilizes homophily scores for categories of organizations to
develop an attribute-based model of structuration weighted by overlapping same category ties,
which is displayed in Figure 4-6. To examine how the information curation behavior of users
following few community organizations may differ from the curation behaviors of citizens who
follow many, low (following only 2 community organization), moderate (3 to 9), high (10 to 49),
131
and extreme (50+) followers are compared. Not shown in Figure 4-6 are unique users, those
following only one community organization, as this visualization would not display any links to
these users using this method. Unique users are still represented, however, in the number of
followers for each organization (represented by node size).
Figure 4-5. Sociogram of 1-mode affiliation matrix of local organizations (colored by
organizational categories).
Figure 4-6 immediately shows media organizations as the category of organization with
the most followers among all users. However, media holds comparatively weak ties with other
organization types. Among all categories of community organizations, however, media accounts
are more frequently co-followed among citizens that follow citizens’ associations. This indicates
that citizens that follow community media organizations- local television stations, newspapers, and
online news organizations- do not often follow other types of community organizations on Twitter.
This tendency changes, however, among high and extreme users who co-follow organizations
across most categories.
132
Entertainment organizations hold similarly weak ties in this model across all categories of
organizations, and this tends to hold among more embedded users- active and extreme users
following many and diverse organizations. These patterns possibly suggest that entertainment
organizations, including a civic center that hosts concerts and sports events, and a golf course and
country club, might be more often followed by non-local users with less connections to the
community.
Figure 4-6. Attribute-based model of structuration of community assets (labels sized by number of
followers; ties weighted by co-followership. Upper left shows structuration of assets among citizens
following 2 organizations; upper right shows 3 to 9; Lower left shows 10 to 49; Lower right shows
50 and over).
The organization type that is most frequently co-followed with other community
organizations is citizens’ associations, which shares strong ties with all other organization types
133
except for entertainment. These ties hold across follower type, meaning that even users following
few organizations tend to curate at least one civic association in their social network. Of note are
followers of schools who, even among users following very few organizations, hold especially
strong ties with citizens’ associations.
In a comparative analysis among follower types, Twitter users in the extreme and high
follower groups curate similar social networks, which feature co-following among most
organizational categories in the community. In contrast, low and moderate users curate information
infrastructures that center on citizens associations that are co-followed with most other categories,
including relatively strong ties with bars and emergency services.
Findings: Collecting Hyperlocal Social Media Data
At approximately 2pm on May 1st, 2017, the National Weather Service (NWS) issued
Tornado Watch Number 185: “A fast-moving line of storms is expected to progress across parts of
New York and Pennsylvania into this evening. Damaging wind will be the primary hazard, with a
few tornadoes also possible” (NWS, 2017). Over the next few hours, Twitter activity marked the
eastward progress of the storm as it approached and then struck communities in Centre County
(Figure 4-7). Tweets warning of the storm and possible tornadoes spike after the 2pm NWS notice,
followed by a flurry of weather forecasts at 5pm anticipating the impact of the storm.
The peak of the storm occurs approximately 20 minutes after 6pm with a sudden downpour
of rain and wind gusts reaching over 60 mph. In a small community in the east of the county, an
EF1 tornado touched down damaging several buildings, severing power lines, and uprooting trees
over a one mile path (NWS, 2017). For other communities, severe winds downed trees blocking
roads and damaging buildings, while heavy rains caused flooding throughout the county.
134
Immediately following the impact of the storm, reports of damage as well as power and Internet
outages instantly spiked. Near 7pm the skies cleared rapidly to reveal a suddenly calm and beautiful
sky.
Over the course of the storm, Centre County 911 would process over 500 calls. Most callers
reported damage sustained from downed trees, including fires started from trees fallen on power
lines. More than 12,000 people lost power, causing the local power company to call in utilities
crews from neighboring areas, and the activation of the emergency operations center to notify
electricity and telecommunications repair crews of areas reporting outages (Annarelli & Hartley,
2017).
Figure 4-7. Total on-topic tweets (blue) and situational report tweets (red) in the Location,
Keyword, and Network Datasets collected on May 1st.
Distribution of Situational Reports
If public safety officials were monitoring Twitter on May 1st using existing data collection
methods, they would identify less than half of all situational reports of infrastructure damage and
0
50
100
150
200
250
300
350
400
12pm 1pm 2pm 3pm 4pm 5pm 6pm 7pm 8pm 9pm 10pm 11pm
Location Dataset Keyword Dataset Network DatasetLocation Dataset Keyword Dataset Network Dataset
135
service disruption. Location and keyword filtering identify 28% (n=97) and 20% (n=72),
respectively, of situational reports collected across the three methods (n=352). In contrast, network
filtering identifies over half, 52% (n=183), of all situational reports posted during the storm (Table
4-5).
After removing overlaps, tweets collected by more than one method, network filtering
identifies 56% (n=119) of unique situational reports. Among the three methods, keyword filtering
returns the least unique data, responsible for 38% (n=27) of all situational reports. As location and
keyword filtering remain de facto standards for real-time data collection, our empirical findings
suggest that existing methods used to establish situational awareness during a crisis remain severely
limited.
Table 4-5. Situational reports collected in the Location, Keyword, and Network Datasets.
Location Keyword Network Total
Power Line 2 6 6 14
Property 17 12 51 80
Road 11 18 26 55
Storm 14 13 14 41
Internet 4 - 4 8
Electricity 49 23 82 154
Total 97 (28%) 72 (20%) 183 (52%) 352 (100%)
We performed a one-way ANOVA to assess if the types of situational reports (e.g. property
damage, electricity outage, etc.) vary in frequency across the three data collection methods. We
found no statistically significant difference between the types of situational reports collected by
each method (F(2,15) = 1.311, p > 0.1). Furthermore, a Tukey post hoc test revealed no significant
differences, with each grouping following the same ratio (Location-Keyword, p=0.935; Network-
Keyword, p=0.299; Network-Location, p=0.474). Thus, while location, keyword, and network
filtering collect different data, and in different volumes, each method tends to collect the same types
of situational reports.
136
Importantly, the three methods each collect different data. While this might be expected
when submitting three different queries to Twitter’s Streaming API, the large number of unique
tweets returned by each method demonstrates the diversity, and volume, of information users post
on social media during a crisis. Overlaps occur across all three datasets, but in very small numbers
(Figure 4-8). For example, while location and network filtering collected a total of 17,295 and
16,733 unique tweets, respectively, only 530 tweets were collected by both methods. Only 10
tweets of the 44,655 total tweets were collected by all three methods.
Interestingly, however, overlaps are much more likely to provide relevant information than
unique data returned by a single collection method (Figure 4-8). For example, 5% of tweets
overlapping the Location and Network Datasets, and 28% of tweets overlapping the Keyword and
Location Datasets, provide situational reports. Relatedly, we find that users who posted situational
reports provided more geographic information than other users, including those discussing the
storm (i.e. users posting on-topic tweets). In comparison to approaches that combine real-time and
post-hoc data collection, such as methods collecting the entire tweet stream of users first identified
using real-time location and keyword-filtering methods (Leysia Palen & Anderson, 2016), we find
that the vast majority of all tweets in the Location (95%), Keyword (98%), and Network (99%)
datasets are posted by users not otherwise identifiable in another dataset. However, among
situational reports the proportions of tweets posted by unique users to the Location (58%), Keyword
(32%), and Network (41%) datasets drastically decrease. This means that users who posted
situational reports more often included multiple types of geographic information- for instance, by
geotagging their tweets and following organizations located in the county- in tweet(s) posted during
the storm than users who posted other types of information.
137
Figure 4-8. Unique and overlapping tweets in the Location, Keyword, and Network Datasets.
Mapping Situational Reports
Observing that location, keyword, and network filtering collect different data in different
volumes, we also analyzed the distribution of situational reports identified by each method across
the 44 extant incidents- physical events of damage or disruption- Twitter users collectively reported
during the May 1st storm. Mapping situational reports collected by each method to the geographic
location of the incidents they report demonstrates the bounds of observation scoped by each data
collection method and suggests how each affords different opportunities for situational awareness
during a crisis.
Each circle in Figure 4-9 represents an incident during the storm, with the circle’s radius
indicating the number of tweets reporting the incident, and color indicating the incident type (Figure
4-9). Over half of the incidents (n=25) are reported by multiple tweets, with the flooding of a high
school football stadium reported most often among tweets in the Situational Reports Dataset
(n=21). Of 44 total incidents reported by Twitter users, network filtering identifies 73% (n=32),
while location and keyword filtering both identify 43% (n=19) each. However, all three methods
identify incidents not reported in another dataset, with unique incidents identified by location (n=5),
138
keyword (n=6), and network filtering (n=10) collectively accounting for nearly half, or 48%, of all
incidents reported on Twitter.
Figure 4-9. Incidents identified by situational reports in the Location, Keyword, and Network
Datasets: power line damage (teal), property damage (yellow), road damage (pink), storm (orange),
internet outage (red), and power outage (blue).
Users posting situational reports during the storm describe the location of an incident using
either a street name or local landmark, but rarely both. For example, all 21 reports of flooding at
the high school football stadium refer to specific (i.e. Memorial Stadium) or general landmarks
(e.g. football field). Of the 44 total incidents, 45% (n=20) were described using only street
information (occasionally including street numbers), while 43% (n=19) were described using only
local landmarks (e.g. names of buildings, businesses, neighborhoods, etc.). In contrast, users
provided both street and landmark information for only five of the 44 incidents. Geographically,
139
incidents identified through each method demonstrate a similar pattern: situational reports are
concentrated within the largest city in the county, State College, with only scattered reports in less
populous communities. This distribution can be expected given the different populations of
communities across the county, but also recalls prior studies that find concentrated reporting around
visible incidents in urban centers while less reporting in peripheral areas that may potentially
experience more damage.
In this respect, comparing the most reported incidents- flooding of the high school football
stadium (n=21), downed trees in a busy intersection (n=13), and downed power lines across a major
roadway (n=10)- with the least reported- unique incidents identified by a single tweet- provides
insight into the information behaviors of social media users reporting events during an emergency.
While the three most reported incidents occur in highly frequented areas in the largest city in the
county, they also are first reported by an influential social media account and subsequently reported
by other, less influential users who may be providing derivative or non-eyewitness accounts. The
stadium flooding was first reported by the popular Penn State University news site, Onward State,
while the incidents of roadway obstruction were first reported by a local news reporter and
meteorologist, respectively. These tweets mentioned other influential accounts (e.g. the local news
station) and included established, highly-visible hashtags (e.g. pawx). In contrast, unique incidents
were often reported by personal accounts lacking mentions and hashtags (e.g. “No Power on 700
Block of Bishop St. Damn this Sucks”). These patterns point to generative and derivative
information behaviors among social media users during a crisis: some users post first-hand accounts
of events while other, geographically remote users share, modify, or discuss this information
(Starbird et al., 2010). We further discuss the methodological and theoretical implications of these
findings in the next section.
140
Discussion
For emergency responders using social media to detect situational reports of emergency in
a geographic area requires new methods for collecting hyperlocal social media data. In the case of
Twitter, emergency responders are limited to the 1-3% of tweets with geographic metadata (i.e.
geotags) and tweets containing keywords (e.g. “flood”) selected to filter the global Twitter stream.
While location filtering excludes most tweets posted to Twitter, keyword filtering excludes any
tweet lacking the narrow set of geographic (e.g. city names) or emergency-related keywords (e.g.
flood) that can make information visible during a global Twitter search. In contrast, Social
Triangulation and network filtering supplement these methods by identifying hyperlocal social
media users and collecting hyperlocal social media data, respectively. In a study comparing social
media data collected using all three filtering methods during a severe weather emergency, network
filtering enables emergency responders to identify twice as many situational reports of
infrastructure damage and service disruption as location and keyword filtering combined.
The complementary methods of Social Triangulation and network filtering present two
main contributions for collecting social media data for community situational awareness. First, by
inferring networks of local social media users and collecting data they create during an emergency,
the methods increase the volume hyperlocal social media data available to emergency responders
and, in turn, can expand awareness of hyperlocal events reported on social media. In this regard,
network filtering complements location and keyword filtering as data collection methods that
should be used together. In the case examining situational reports in State College during a severe
weather emergency, each of the three methods identify unique incidents of infrastructure damage
and service disruption reported on Twitter, but network filtering alone identifies nearly three
quarters (73%) of all incidents reported during the emergency. These findings suggest that
combining multiple data collection methods is necessary when using Twitter to support situational
141
awareness during a crisis. Furthermore, by inferring local social media users as well as their sources
of community information, Social Triangulation can inform emergency communications planning
by allowing emergency responders to identify “community influencers” positioned to re-distribute
official messages to the widest audience possible in a community. Consequently, this study
discusses emergent guidelines for employing social media in emergency communications planning:
account for local “filter bubbles,” identify community influencers, and cooperate with citizens’
associations.
Identifying Hyperlocal Social Media Users and Information Sources
This exploratory study finds evidence supporting the underlying assumption of ST: that
local social media users tend to curate their social networks around community organizations in
their geographic area. Among the 79,998 users following at least one organization in the city
situating this study and with self-identified location information, fully 68%, or 54,165, identify
themselves as locals. Moreover, the more community organizations a user follows the more likely
they are to self-identify as local. For unique and low followers who curate only one or two
community organizations in their social network, 44,885 or 67% identify as local. This increases
for moderate (71%) and high (84%) followers. Among people who follow 50 or more community
organizations, fully 98% identify as local. This indicates that the more local organizations a Twitter
user follows, the more likely they identify themselves as local to that community.
In addition, this study finds that locals following community organizations differ with
respect to their level and position of embeddedness within the community information
infrastructure. Local media accounts on Twitter, including television, radio, and news
organizations, possess the most followers among the categories of community organizations
142
catalogued. However, the majority of users following only one or two community organizations on
Twitter (160k+) disproportionately curate media organizations within their social networks. In this
respect, the majority of locals remain weakly embedded within the community information
infrastructure.
In contrast, a sizeable minority of users (20k+) curate their social networks to follow
multiple and various kinds of community organizations on Twitter, including media accounts, the
most followed organizations in the community, as well as more citizens’ associations and civic
organizations compared to the majority of weakly embedded users. In this respect, moderate, high,
and extreme followers, those users following multiple community organizations, emerge as
strongly embedded in the community information infrastructure, uniquely positioned to access
certain kinds of information disseminated in the community.
Guidelines for Emergency Communications Planning
Together these findings suggest that ST can inform emergency preparedness by identifying
hyperlocal social media users and the community organizations that serve as important (or
potential) sources of hyperlocal information. Community preparedness constitutes an important
component of community resilience, the ability of a community to withstand and recover from a
disaster or emergency (Chandra et al., 2011). Community preparedness involves capacity-building
efforts undertaken before a crisis that result in more resilient communities. Attention to
preparedness and resilience, balances the reactive posture of local emergency management work
traditionally concerned with crisis response (Cutter et al., 2013), with proactive and integrative
efforts to involve both officials and everyday citizens in forms of community capacity-building
(Hernantes, Labaka, Turoff, Hiltz, & Bañuls, 2017). If deployed in support of emergency
143
communications planning, an ST approach could contribute to community emergency
communication before, during, and after emergencies. The ST approach outlined here is likely
feasible for most communities: ST requires motivated volunteers, basic expertise interacting with
Twitter’s public API, and limited computing resources.
Previous studies find that municipal governments and emergency managers lack strategies
and best practices for using social media during emergencies (Hiltz & Plotnick, 2016; Hughes et
al., 2014), as well as guidelines for using social media for emergency communications (Houston et
al., 2015; Rice & Spence, 2016). While general guidelines for social media use have been outlined
and evaluated (Brynielsson et al., 2018; Kaufhold, Gizikis, Reuter, Habdank, & Grinko, 2018;
Wukich, 2018), strategies for emergency communications planning with social media remain
lacking. As ST reveals the social media accounts of local users and local information sources, ST
can help officials understand who uses social media in their community and how to reach them.
Based on the findings of this study, emergent guidelines for emergency communications planning
with social media are presented below.
1) Identify Filter Bubbles
Emergency communications planning with social media must account for an uneven social
media landscape and the “filter bubbles” of users curating limited or similar sources of community
information within their social networks (Pariser, 2011). This study finds the information
infrastructure of the community examined to be highly uneven, with followers of community
organizations embedded at different positions and levels of the network. This suggests Public
Information Officers (PIOs) or any officials attempting to use social media to communicate with
the public will likely encounter a minority of local users positioned to immediately access
information shared from a variety of official accounts, while information access among the majority
144
of users will depend on numerous non-official accounts re-distributing official messages to
fragmented networks of local users. Consequently, as the dissemination of time-critical information
stands to proceed unevenly throughout a community, dissemination strategies must be deliberately
planned and coordinated in advance of emergency situations.
Importantly, this study shows that 139,440 or 75% of all user follow only one of the 195
organizations catalogued in the city. While unique users typically follow the social media account
of a media organization, each category of community organization features a significant following
of unique users who, among those evaluated, tend to identify as locals. This majority of users, as
well as the many users who follow multiple organizations of the same category (e.g. only bars),
indicate the presence of “filter bubbles” or populations of users who have self-selected, through the
curation of their social media networks, to receive limited or similar information. ST identifies
filter bubbles associated with each category of community organization, significant audiences for
whom dissemination strategies can be tailored in emergency communications planning.
2) Identify Community Influencers
Given the limited reach of official social media accounts and the presence of multiple “filter
bubbles” among local users, effective emergency communication using social media will likely
require cooperation between officials and “community influencers”: citizens and organizations
with many social media followers in the community. Given the unevenness of community
networks, effective dissemination strategies require cooperative practices of re-distribution
between a message source, such as the social media account of the local emergency management
agency, and community influencers positioned to re-distribute official messages to both wide and
targeted audiences (e.g. filter bubbles). Civic and emergency response officials charged with
145
emergency communications planning might do more than assume their messages reach their target
audiences by identifying and seeking the participation of community influencers.
For the city in which ST was deployed, the reach of emergency and civic services’ social
media accounts, by number of followers, remains limited compared to many community
organizations. While the Twitter accounts of emergency services are collectively followed by 16k
users, the two local television stations and local newspaper are followed by over 15k users each.
More dramatically, a local online newspaper has over 100k followers. Officials conducting
emergency communications planning might seek arrangements with these organizations to retweet
official information in the event of an emergency. Such arrangements might involve forms of
coordination between the social media managers of these organizations and PIOs serving in
municipal and emergency management agencies (Hughes, & Palen, 2012). Coordination efforts
might involve establishing a community hashtag (Rice & Spence, 2016; Starbird & Palen, 2011),
and/or manual or automatic retweeting between accounts in certain emergency situations. This
study observed sporadic retweeting among diverse community accounts and emergency services
(e.g. bars retweeting police posts soliciting information on wanted persons). While outside the
scope of this study, these ad hoc, cooperative practices indicate a shared interest in community
awareness that suggests opportunities for deliberate planning and coordination.
Consequently, this analysis suggests the importance of involving citizens and organizations
positioned as community influencers in emergency communications planning processes to
deliberately reach the fragmented networks that comprise the information infrastructure of a
community and supplement the ad hoc, emergent diffusion of information that occurs on social
media during an emergency. By mapping the information infrastructure of a community, ST
enables the identification of community influencers and, in turn, opens opportunities for official-
citizen cooperation that can enhance emergency communications during a crisis.
146
3) Cooperate with Citizens’ Associations
Citizens’ associations represent community information bridges, connecting the follower
networks of different categories of community organizations. This analysis consistently observed
co-following between citizens’ associations- nonprofit and volunteer community groups such as
animal rescue shelters, conservation societies, food pantries, etc.- and other categories of
community organizations. These curation behaviors indicate that users who follow citizens’
associations are most likely to follow other types of community organizations, regardless of the
number of organizations they follow in the community (provided they follow more than one). This
curation behavior suggests potential roles for citizens’ associations and their networks of followers.
Emergency communications planners might cooperate with citizens’ associations and their
followers as these accounts represent seed points for re-disseminating information to citizens with
diverse community connections who, in turn, represent likely information bridges to different
audiences of local social media users. That is, together with the wide reach of community
influencers such as media organizations, citizens’ associations might be able to disseminate
information to local users best positioned to disseminate information across filter bubbles in the
community. This possibility, of course, rests on the reach and composition of these followers’ social
networks and their willingness to re-distribute official information. This assumption, however, may
be more likely given the potential characteristics of citizens’ association followers.
People who follow citizens’ associations at the very least have selected to receive
information from these organizations and, likely, consist of members of these organizations. As
such, these followers- especially those moderate, high, and extreme followers who tend to follow
many more citizens’ associations, civic and emergency services accounts than low and unique
users- may represent active citizens in the community who exhibit forms of social capital that can
147
contribute to community resilience (Aldrich & Meyer, 2015). As such, these citizens
simultaneously occupy embedded positions within both the information and social infrastructure
of the community. As forms of citizen participation and citizen-official cooperation become more
central to “Whole Community” emergency management work (FEMA, 2011; Schafer, Carroll,
Haynes, & Abrams, 2008), these citizens stand to play active, overlapping roles not only as re-
distributors of information but, potentially, in community-centric variants of remote, digital
volunteering that play important roles in disaster response (Hughes, & Tapia, 2015; Reuter, Heger,
& Pipek, 2013).
While this discussion of emergency communications planning focuses on efforts of
emergency response officials to reach local users, it remains important to recognize that such
communications planning cannot sufficiently account for community preparedness and response
efforts. Social networks among local citizens remain the primary information channels on social
media platforms. Moreover, prominent individuals can be more influential than organizations in
the creation and dissemination of information on social media. The offered guidelines and
discussion are intended to complement and support forms of many-to-many citizen communication
that underpin any contribution of social media to community preparedness and resilience.
Guidelines for Geo-Inference System Design
This study introduced ST as a procedure for identifying locals and mapping the information
infrastructure of a community by identifying sources of community information and their
audiences. ST draws on external, community-based information to identify these sources, and infers
the location and describes the information curation behaviors of local citizens who access
information these sources provide. In the tradition of community asset mapping, and in lieu of
148
available geo-inference systems, ST provides a manual procedure for community preparedness
work that communities can do right now, without waiting for a future system that they may never
afford or have the skills to create. ST, however, is labor-intensive, requiring people to manually
catalogue hundreds of community social media accounts. As a result, manual processes of ST might
not scale to populous, urban areas.
However, the generalizable procedure for community asset mapping and analysis that ST
provides also suggests guidelines for the design of automated geo-inference systems. These
complement prior geolocation inference methods that infer the location of a user according to his
or her network ties by insisting on the importance of ties users develop with hyperlocal information
sources. The first and second phases of ST involve categorizing and cataloguing sources of
community information constrained to activities and events in a specific, geographically-bounded
space. This manual procedure requires external sources of information that can identify and
catalogue resources as hyperlocal. This requires differentiating constrained from unconstrained
information sources whose scope of coverage extends beyond the exclusively local. The New York
Times, for instance, would not meet the hyperlocal criterion as it provides coverage of events both
in New York City and internationally, and has a correspondingly international audience. In contrast,
a local newspaper that reports on events only within a community can be suspected to have a
primarily local audience of readers.
This intuition, the basic premise of ST, has been observed by McGee et al. (2013), who
find that predicting the location of a user can be effectively accomplished by looking within their
social network for friends/organizations and their contacts (e.g. followers) who all live near each
other (for friends/contacts whose location data is available). The authors measure the Local Contact
Ratio (LCR) of users’ friends, the fraction of a friends’ contacts who live within 25 miles of the
149
friend. Twitter accounts of local newspapers often emerge as a significant predictor of an unknown
user’s location according to this measure. McGee et al. write:
a local newspaper may have thousands of followers and few friends, but the people who
follow a newspaper are generally local. According to the other factors we looked at [e.g.
reciprocal following/mentions, influence, etc.], the newspaper is a bad predictor of
location, but in reality it is a great predictor… this is the one that is most strongly correlated
with distance. (p. 466)
From this perspective, the sources of community information ST attempts to catalogue would be
those with a high LCR. Accounts with a high LCR in a community would, therefore, be most useful
in inferring the location of citizens in the community who follow these accounts but lack individual
geolocation data.
Adopting ST for the design of a (semi-)automated geo-inference system might therefore
leverage LCR as a measure of community information to identify a list of accounts that an end user
could then categorize and catalogue as community information sources when performing ST. Such
a system could draw on tweets geotagged within a single community to identify a set of seed users
located in that community. Extracting the accounts followed by these seed users as a provisional
set of community information sources, the followers of these accounts that include location data
could next be extracted to measure the LCR for each provisional source of community information.
Thus, for each provisional source with a known location (e.g. profile information or geotags) an
LCR measure and rank (e.g. number of followers) could be determined and presented to an end
user who could then categorize (e.g. civic service, media, etc.) and catalogue these accounts as
community information sources. For global or national-level geolocation inference methods this
approach remains too computationally expensive due to the volume of data and rate limiting by
Twitter’ API when considering the size of a network including seed users, source accounts seed
users follow, and the followers of source accounts (McGee et al., 2013; Jurgens et al., 2015).
However, as related research observes (J. Zhang, Sun, Zhang, & Zhang, 2015), these costs can be
150
mitigated if employed for a single community. Automating ST is likely to be similarly feasible for
many communities that can employ citizens with IT experience and modest computational
resources.
Collecting Hyperlocal Social Media Data
This study introduces a novel network filtering method to collect Twitter data during a
crisis and, in the context of a hyperlocal weather emergency, comparatively assesses how location,
keyword, and network filtering methods can enhance situational awareness. In doing so we present
two contributions to crisis informatics research examining the relationship between data collection
methods and opportunities for situational awareness during a crisis.
First, we introduce network filtering as a novel data collection method and empirically
demonstrate how network filtering can dramatically increase the ability to collect data supporting
situational awareness during a crisis. In the case of a severe storm we find that network filtering
doubles the amount of on-topic, weather-related information, as well as situational reports of
infrastructure damage and service disruption, collected from Twitter compared to location and
keyword filtering methods. Conversely, these findings suggest that situational awareness
technologies employing typical location and keyword-based data collection methods overlook a
significant amount of relevant information during a crisis.
This study also reveals that location, keyword, and network filtering all provide unique
opportunities for situational awareness. That is, each method collects different data in different
amounts, including unique data providing unique insights into incidents occurring across a
geographic community. We find that nearly half (48%) of all incidents reported on Twitter during
the May 1st storm can be identified only by combining all three data collection methods.
151
Furthermore, tweets collected by multiple collection methods are more likely to provide situational
information than tweets collected by a single method alone, suggesting potential filtering strategies
that can reduce dataset size and noise. By introducing network filtering as an effective data
collection technique and recommending the pairing of multiple data collection methods- location,
keyword, and network filtering- to expand and scope data collection during a crisis, this study
makes an important methodological contribution to the design of situational awareness tools that
can expand awareness during times of crisis.
Second, this study contributes to our understanding of crisis information behavior by
suggesting that the types of situational information users report on social media are shaped by
highly-visible information posted by influential social media accounts. Recalling the distinction
between generative, eyewitness reports and derivative reports posted by social media users
(Starbird et al., 2010), our analysis illustrates how influential social media accounts can distribute
reports of events during an emergency, that, in turn, become topics of discussion among other social
media users in a geographic community. Importantly, this finding provides insight into urban
reporting biases observed among social media users in prior studies (Hecht & Stephens, 2014;
Malik et al., 2015; Shelton et al., 2014). As observed here, the most reported incidents on Twitter
were those early reported by influential accounts in the community and subsequently reported by
others. This finding suggests that urban reporting biases result both from the demographics of social
media users posting geotagged situational information (e.g. younger, more urban, etc.), as well as
derivative information behaviors shaped by popular social media accounts that influence what
information becomes visible and discussed among those social media users.
In addition, our findings provide further evidence that local social media users often omit
the types of course-grained geographic information (e.g. city names) that makes tweets visible to
remote Twitter users and more easily collected using keyword-based methods (Saleem et al., 2014;
152
Vieweg et al., 2010). However, we find that users posting situational reports of infrastructure
damage and service disruption do include geographic information by naming local streets and
landmarks in their tweets. That users often include one or the other, even when multiple users report
the same incident, suggests social media users in a geographic area tend to share local knowledge,
including standard place names, and communicate with others possessing the same local
knowledge. Importantly, the tendency of social media users to include the names of local streets
and landmarks when posting situational information recommends the creation and use of local
gazetteers when designing situational awareness tools to geolocate situational reports posted during
a crisis (Middleton, Middleton, & Modafferi, 2014).
Limitations and Future Work
Important limitations characterize this introduction of ST and point to directions for future
work. First, and most significantly, this analysis only recognizes community organizations as
sources and disseminators of community information. Of course, this remains far from the case.
Everyday citizens drive the social media information cycle and constitute not only passive
recipients of emergency-related information disseminated by emergency management officials but
information producers and primary participants in the mitigation of, preparation for, response to,
and recovery from small emergencies to large-scale crises. The focus on community organizations
represents only an exploratory, and necessarily limited application of ST. The concept of ST
presents a flexible approach for cataloguing social media accounts providing community
information as a geographic “ground truth” in relation to which the location of users following
these accounts can be inferred and evaluated.
153
Second, while this study explores ST as a community project necessarily situated in a
specific place- a city in the Northeastern United States- insights can be gained by deploying ST in
different geographic and cultural contexts (Huang, Wu, & Cheng, 2016). The scope of this study
provides for general procedures and guidelines for deploying ST that can be taken up by different
communities. These procedures provide a flexible and efficient approach to identifying and
cataloguing a ground truth dataset for inferring and analyzing a community information
infrastructure. Future work might compare the data provided by ST with other data collected using
geotags and keyword to examine how this data may be similar and different. The results of this
extension of ST may hold powerful implications for how data aggregation methods may be
combined to increase what is known about geographic communities in times of crisis. Further, the
ST results of several communities may be used to compare the information infrastructure of several
communities as a method to learn more about how filter bubbles impact local communities
differently. Such an analysis would further examine the relationship between user information
curation behavior and geographic location.
Third, the effort to ascertain users’ locations using available profile information requires
further methods of evaluation. An immediate opportunity presents itself in the collection and
analysis of geotagged tweets posted by users following local organizations. In this regard, users
with geotagged posts might be identified as local citizens according to three separate localness
metrics defined by Johnson and colleagues (2016): if the user posts tweets in the municipal locale
at least n-days apart (n-days metric); if the majority of the user’s tweets occur in the locale (plurality
metric); or if the geographic median of all the user’s tweets falls within the locale. These metrics
could be compared among and between users following the same number of organizations as well
as categories of organizations. Again, such an analysis could further reveal of the relationship
between user information curation behavior and geographic location.
154
Lastly, when employing ST and network filtering for situational awareness possible
limitations may arise from our analysis of an emergency of relatively small scale and its difference
from disasters affecting larger populations and geographic areas (Olteanu et al., 2015). While we
find similar types of information reported by social media users, the method of network filtering
we have introduced would likely require the incorporation of automated techniques to infer and
collect data from the more expansive networks of users in areas affected by a disaster.
Conclusion
Social Triangulation presents a novel method for inferring local users vis-à-vis the local
organizations they follow on social media. Based on the hypothesis that local citizens are more
likely to follow the Twitter accounts of local organizations in their communities than non-local
citizens, Social Triangulation involves cataloguing and categorizing the accounts of local
organizations to infer community networks among Twitter users who tend to follow multiple, local
organizational accounts. The method is then evaluated using users’ self-identified profile locations.
Among 79,978 Twitter users following a local organization in State College and including location
information in their profiles, 54,165 (68%) self-identify as living in State College. Moreover, the
more local organizations users follow the more likely they are to self-identify as local citizens. For
users following one or two local organizations, 37,788 or 67% identify as local citizens. This
increases among those who follow 3-9 organizations (71%), 10-49 organizations (84%), and,
among those following 50 or more organizations, 98% self-identify as local. In addition, by
categorizing the Twitter accounts of local organizations, Social Triangulation identifies important
sources and types of community information followed among local Twitter users. In the case of
State College, local media accounts reach the most users, however distinct “filter bubbles” exist
155
among networks of users following a single category of organizational account(s) (e.g. schools,
civic services, bars, etc.).
Second, network filtering offers a complement to existing location and keyword filtering
data collection methods. Network Filtering collects social media data from networks of Twitter
users identified through Social Triangulation during an emergency. This study analyzes the
distribution of situational reports of infrastructure damage and service disruption across location,
keyword, and network-filtered social media data during a severe weather emergency which saw the
touchdown of an F1 tornado in State College, Pennsylvania. Findings reveal that location and
keyword filtering respectively identify 28% (n=97) and 20% (n=72) of all situational reports
collected across the three methods (n=352). In contrast, network filtering doubles, 52% (n=183),
the number of situational reports collected in real-time compared to location and keyword filtering
alone, suggesting that existing methods to collect social media data in support of providing
situational awareness during a crisis remain severely limited.
The complementary methods of Social Triangulation and network filtering present two
main contributions for collecting social media data for community situational awareness. First, by
inferring networks of local social media users and collecting data they create during an emergency,
the methods increase the volume hyperlocal social media data available to emergency responders
and, in turn, can expand awareness of hyperlocal events reported on social media. In this regard,
network filtering complements location and keyword filtering as data collection methods that
should be used together. In the case examining situational reports in State College during a severe
weather emergency, each of the three methods identify unique incidents of infrastructure damage
and service disruption reported on Twitter, but network filtering alone identifies nearly three
quarters (73%) of all incidents reported during the emergency. These findings suggest that
156
combining multiple data collection methods is necessary when using Twitter to support situational
awareness during a crisis.
Second, by inferring local social media users as well as their sources of community
information, Social Triangulation can inform emergency communications planning by allowing
emergency responders to identify “community influencers” positioned to re-distribute official
messages to the widest audience possible in a community. Furthermore, this study presents general
guidelines for employing social media in emergency communications planning: identify filter
bubbles, identify community influencers, and cooperate with citizens’ associations.
157
Chapter 5
Phase III: Integration
As emergency dispatch centers such as PSAPs already coordinate the flow of information
between citizens and first responders, incorporating social media requires 911 call takers,
dispatchers, and prospective social media analysts to make sense of information gathered from
social media using information gathered from emergency callers, and vice versa. In this regard,
incorporating social media analytics in emergency dispatch continues the transition of PSAPs from
reactive call centers to proactive data analytics and coordination hubs using Next-Generation 911
infrastructure to collect and process heterogenous data from physical and social sensors to support
the situational awareness needs of first responders. Phase III describes role plays and simulations
conducted with 911 call takers, dispatchers, and social media analysts to examine how future
emergency dispatch workflows can integrate information from social media and 911 calls during
an emergency. These studies suggest sociotechnical requirements for including social media
analysts and analysis tools in next-generation emergency dispatch work.
Literature Review: Evolution of Next-Generation Emergency Dispatch
When someone dials an emergency telephone number like 9112, they begin an information
processing task that culminates when a first responder arrives at the scene of an emergency. While
information dispatched to first responders- fire, medical, police- constitute the output of this task,
the inputs, such as provided by 911 callers, are increasing in volume and variety. The development
2Or 1-1-2, 1-1-9, 9-9-9, etc. This study involves emergency dispatch in the United States and will therefore
refer to 911, the emergency telephone number in North America.
158
of Next-Generation 911 systems allow people to call, text, and use web-based applications to
contact emergency dispatch centers channels (Holland, 2018), known as Public-Safety Answering-
Points (PSAPs), which, increasingly, also seek to detect emergencies and gather situational
awareness information using data collected from physical and social sensors, including social
media.
In this data-rich environment, the work of emergency telecommunicators coordinating
information flows between citizens and first responders- the throughput of emergency response-
becomes much more complex. Anticipated by data analytics in policing beginning with early
statistical tools such as CompStat in New York City and, more recently, dedicated Strategic
Decision Support Centers in Chicago (Smith, 2018), PSAPs are beginning to process an array of
data inputs from physical (e.g. cameras, license plate readers, gunshot detectors) and social (e.g. 3-
1-1, 911 calls, social media) sensors and digital records (e.g. arrests, complaints, summonses,
firearm registries) (Levine & Tisch, 2014). Consequently, PSAPs are transforming from reactive
call centers to proactive data analytics and coordination hubs providing first responders with real-
time information before and during emergencies.
In this regard, incorporating social media analytics presents emergency dispatch centers
with distinct opportunities. PSAPs can use social media to disseminate warnings, public notices,
and other emergency-related information, especially during crises when call processing may be
disrupted or delayed due to high-call volumes. Utilized as a distributed reporting system, social
media can alert PSAP staff to developing emergencies and augment situational awareness
information provided by 911 callers (Grace et al., 2018). For PSAPs, opportunities to incorporate
social media require, first, developing analytic software that provide dispatch centers with new
information inputs and, second, information obtained from social media data within the throughput
of actual emergency dispatch work (Grace et al., 2018).
159
However, research on automated techniques to collect and process (Imran et al., 2015;
Olteanu et al., 2014) social media data continue to outpace efforts to incorporate social media
analytics within the distributed workflows of telecommunicators working in emergency dispatch
centers: call takers and dispatchers who assist and gather information from 911 callers and provide
first responders with situational awareness information, respectively (Grace et al., 2018). The latter
requires integrating information obtained from social media with information obtained from
existing data sources, notably 911 callers, before the inputs of social media can be judged relevant,
useful, or “actionable” in emergency dispatch work (Kropczynski et al., 2018; Zade et al., 2018).
Similar to humanitarian responders, emergency managers, and Public Information Officers (PIOs)
(Dailey & Starbird, 2017; Hiltz & Plotnick, 2016; Hughes, & Palen, 2012; Zade et al., 2018),
telecommunicators make sense of social media in contexts of emergency dispatch work, and with
respect to evolving information requirements and other, often more trusted, sources of information.
In these contexts, information reported on social media can be evaluated and utilized to supplement
available information, filling gaps in situational awareness, or discarded if redundant, imprecise, or
irrelevant (Grace et al., 2018; Kropczynski et al., 2018).
This study investigates the incorporation of social media analytics in emergency dispatch
by examining how telecommunicators integrate information from social media with information
gathered by 911 call takers to enhance information dispatched to first responders (Table 5-1).
Through role play scenarios involving telecommunicators in a busy metropolitan PSAP using
synthetic datasets and an analytic dashboard, this study examines the emergency dispatch
workflows that develop when social media analytics and analysts are introduced in mock scenarios
featuring a mall shooting and severe flooding. The resulting findings i) provide guidance to
emergency dispatch centers incorporating social media analytics, ii) inform the design of automated
methods and analytic dashboards for processing and visualizing social media data, and iii) theorize
160
information integration as a process of distributed sensemaking involving multiple actors using
multiple systems and data sources to understand unfolding events during an emergency.
Table 5-1. Phase III research questions
How can emergency dispatch officials integrate information from social media and 911
callers in emergency dispatch operations?
RQ1 What occasions breakdowns in 911 call taking and dispatch, and how can social media
analysts contribute to the sensemaking efforts they invite?
RQ2 What interpretive frameworks do 911 telecommunicators draw on to identify
situational information, and how can they guide social media analysts?
RQ3
What sociotechnical infrastructures sustain the sensemaking efforts of 911
telecommunicators, and how can they support cooperation with social media
analysts?
RQ4 How do emergency dispatch staff analyze and seek information on social media in
coordination with emergency call taking and dispatch operations?
RQ5 How do emergency dispatch staff integrate information from social media and
emergency callers?
RQ6 How do emergency dispatch staff discover inconsistencies among information from
social media and emergency callers?
RQ7 How do emergency dispatch staff evaluate inconsistent information from social media
and emergency callers as supplemental, unique, or redundant?
Incorporating Social Media in Emergency Dispatch
Prior studies describe emergency dispatch work as intensively cooperative, characterized
by complex information flows that require coordination, and mediated by Computer-Aided
Dispatch (CAD) systems and numerous devices and applications which enable and constrain the
work of distributed citizens, telecommunicators, and first responders (Furniss & Blandford, 2006;
161
Norri-Sederholm, Seppälä, Saranto, & Paakkonen, 2016; Preusse & Gipson, 2016). Focusing on
social media, Boersma and colleagues (2016) examine the trial of Twitcident, a web-based system
for filtering, searching and analyzing social media data, within Dutch real-time intelligence centers
supporting emergency dispatch and response. The trial was “disappointing” as dispatch staff had
difficulty pre-selecting search terms to effectively filter data due to the diverse, colloquial language
of social media posts and different needs of emergency fire, medical, and police services. The lack
of a shared interpretive framework for evaluating filtered search results compounded this difficulty.
Moreover, Twicident relied on the 1% of tweets with geotags as the dispatch center utilized location
information to validate and integrate information from social media with existing sources of
information. As a result, most of the tweets were filtered out and as a result, unavailable to dispatch
staff and first responders. Finally, during the trial Twicident provided only redundant information
rather than new information that could support early warning and response (p. 243-4).
This study addresses and expands upon the issues identified by Boersma and colleagues
(2016) by examining how emergency dispatch staff integrate information from social media and
emergency callers. By approaching information integration as a process of distributed sensemaking,
this study explores how analysts in emergency dispatch centers use social media vis-à-vis 911 call
data to i) select search terms and filter data, ii) interpret and integrate incomplete information (e.g.
without location information) across data sources, and iii) evaluate information as unique,
corroborative, or redundant compared to existing sources of information during an emergency.
Theory: From Situational Awareness to Sensemaking
Situational awareness represents a foundational concept for research exploring information
processing and data fusion, human-centered analytics, and decision-making in data and information
162
rich environments. The common definition of situational awareness comes from Endsley (1995)
who describes a “state of knowledge” concerning the perception of elements in the environment,
comprehension of relations among elements, and projection of these elements’ statuses in the
future. In contrast, Endsley differentiates situational awareness from what she refers to as “situation
assessment,” the “processes used to achieve that state” (Endsley, 1995, p. 36). This study, roughly,
concerns the latter, which has elsewhere been referred to as the process of sensemaking (Greitzer,
Schur, Paget, & Guttromson, 2008; G. Klein, Moon, & Hoffman, 2006; Weick, Sutcliffe, &
Obstfeld, 2005).
If situational awareness concerns a mental representation of elements in an environment,
dots that must be connected, sensemaking concerns “the skill needed to identify what counts as a
dot in the first place” (G. Klein et al., 2006, p. 72). Importantly, sensemaking does not bracket the
“situation” as an independent mental model of the world but considers any situation as a situated
encounter with the world that unfolds through (often skilled) engagement in goal-directed activities
(Heidegger, 2010). As a result, to return to the dot analogy, sensemaking concerns understanding
what dots matter in a given situation when pursuing a given task.
In contrast to a situational awareness perspective, a sensemaking approach poses different
questions (Greitzer et al., 2008, p. 2). These questions are guided by the three, interconnected
aspects of sensemaking.
First, breakdowns in perceived activity prompt sensemaking. As Weick et al. observes:
Explicit efforts at sensemaking tend to occur when the current state of the world is
perceived to be different from the expected state of the world… To make sense of the
disruption, people look first for reasons that will enable them to resume the interrupted
activity and stay in action. These "reasons" are pulled from frameworks such as
institutional constraints, organizational premises, plans, expectations, acceptable
justifications, and traditions inherited from predecessors. (Weick et al., 2005, p. 409)
Breakdowns involve contrast between the expected and encountered course of an activity.
Importantly, the expected course of an activity implies an interpretive background or framework
163
that shapes our experience of the activity and against which interruptions can be experienced
(Heidegger, 2010). Two important consequences emerge: without an effective basis for comparison
and evaluation (i.e. interpretive framework), more information does not necessarily contribute to
better situational awareness (i.e. additional information will not prompt breakdowns and initiate
sensemaking), and automated systems for data fusion and decision-support may obscure
breakdowns that otherwise might appear to human analysts as contrasts between the expected and
unexpected among data or data sources (G. Klein et al., 2006, p. 72).
Whereas a situational awareness perspective begins with an objective environment of
“elements” assumed to be equally noticeable to a decision-maker, a sensemaking perspective
begins with occasions of breakdown that occur when the encountered and expected course of a task
differs and, as a result, elements of the environment become noticeable to the decision-maker who
then reinterprets the situation to resume the task (Greitzer et al., 2008, p. 2). We therefore ask:
(RQ1) what occasions breakdowns in 911 call taking and dispatch, and how can communications
analysts working with social media data contribute to the sensemaking efforts they invite?
Second, sensemaking efforts, in turn, draw on interpretive frameworks to understand what
matters in a situation. As Weick et al. (Weick et al., 2005) describe, sensemaking efforts initiated
by breakdowns in activity draw on interpretive frameworks in the form of social or institutional
norms, past experiences, and protocols. Such protocols reveal what matters in a situation and guide
the resumption of the interrupted activity. Thus, when encountering roadway congestion, for
example, drivers make use of known detours, prior experiences with area traffic, and the legal
framework of driving rules to resume travel to a destination. Whether the median is wide enough
to bypass the congestion, for instance, does not figure as an important or noticeable element of the
environment for many drivers (but may for some in different driving environments).
164
While a situational awareness perspective assesses information that went unnoticed or
unknown by a decision-maker that would have prevented or solved a problem, a sensemaking
approach seeks to understand the frameworks in which the situation was intelligible to the decision
maker at the time, and “help the decision maker understand what matters, see relationships… [and]
decompose complex information into coherent chunks” (Greitzer et al., 2008, p. 2). We therefore
ask: (RQ2) what interpretive frameworks do 911 telecommunicators draw on to identify situational
information, and how can they guide the analysis of social media data?
Third, sensemaking always remains a situated activity, conditioned by local arrangements
of human and technical resources for action (Star & Ruhleder, 1996). In this sense, the interpretive
frameworks that guide sensemaking become deployed in the relationship between a decision-maker
and the environment, such that understanding what matters emerges through situated interactions
with people and things. To refer to the prior traffic example, taking a detour might involve use of
a smartphone and services such as Google Maps or Waze, or calling a friend for directions:
available resources sustaining inquiry. As a result, while a situational awareness perspective
focuses on information inputs, what was or was not available, a sensemaking perspective attempts
to understand the processes of interaction between decision-makers and sociotechnical
infrastructures that shape the availability of information. We therefore ask: (RQ3) what human and
technical infrastructures sustain the sensemaking efforts of 911 telecommunicators, and how can
communications analysts draw on and contribute to these?
From Sensemaking to Distributed Sensemaking
Following Klein and colleagues (2007), we approach sensemaking as “a deliberate effort
to understand events” (p. 114). Figure 5-1 illustrates the Data-Frame Theory of Sensemaking, a
165
recursive process of modifying or replacing an existing understanding of the situation, what Klein
et al. refer to as “frames.” According to Klein et al. (2007), a frame is “an explanatory structure
that defines entities by describing their relationship to other entities” (p. 118). Frames can take the
form of stories, maps, or scripts (p. 118). Stories provide chronological orderings and causal
relationships, while maps provide spatial arrangements from which directions and routes to
destinations or landmarks can be inferred. Scripts outline cooperative roles and coordinate action
by meshing together actors’ activities. Importantly, frames both explain and guide information
seeking: “a frame is a structure for accounting for the data and guiding the search for more data”
(p. 118).
Shown in Figure 5-1, sensemaking develops within a frame (A) providing understanding
of the situation at hand, to include gaps in understanding that can be addressed by seeking further
information. When encountering a new situation, actors can infer frames using only a few data
points referred to as “cues” (Weick et al., 2005) or “anchors” (Gary Klein et al., 2007): “the initial
one or two key data elements we experience sometime serve as anchors for creating an
understanding. These anchors elicit the initial frame, and we use that frame to search for more data
elements” (p. 122). Incorporating social media in emergency dispatch requires understanding the
repertoire of explanatory frameworks (i.e. frames) telecommunicators use to interpret and seek
information on social media, as well as the characteristics of social media data (i.e. cues) that
facilitate analysis and information seeking (McMaster, Baber, & Duffy, 2012). (RQ4) How do
emergency dispatch staff analyze and seek information on social media in coordination with
emergency call taking and dispatch operations?
The existing frame is elaborated (B) when new information is discovered that provides
additional details about the situation. The elaboration loop in Figure 5-1 represents the process of
seeking and discovering information to fill (or identify) information gaps and establish new
166
relationships between newly-discovered and existing attributes characterizing the frame.
Incorporating social media in emergency dispatch requires understanding a distributed process of
elaboration in which call takers, analysts, and dispatchers seek, analyze, and integrate information
from multiple data sources in ways that add to the collective understanding of an emergency. (RQ5)
How do emergency dispatch staff integrate information from social media and emergency callers?
Figure 5-1. Four functions of sensemaking cycle (Gary Klein, Moon, & Hoffman, 2006; Gary Klein et
al., 2007).
Encountering inconsistent data calls into question (C) the existing frame, which can either
be preserved and elaborated (to include discarding anomalous data), or reframed (D) by selecting
or constructing a new frame. These encounters are critical insofar as they reveal flaws in current
understanding. (RQ6) How do emergency dispatch staff discover inconsistencies among
information from social media and emergency callers? When the attributes of an emergency (e.g.
what, where, who) vary among reports from social media and 911 callers, telecommunicators must
compare multiple, incomplete, and inconsistent reports across data sources and determine whether
the reported information provides new information for a known emergency (i.e. elaboration),
provides redundant information for a known emergency (i.e. preservation), or identifies a hitherto
unreported emergency (i.e. reframing. (RQ7) How do emergency dispatch staff evaluate
167
inconsistent information from social media and emergency callers as supplemental, unique, or
redundant?
Methods: Role Play & Simulation
Emergency Dispatch Role Plays
This study reports findings from a design workshop conducted at a PSAP responsible for
911 call taking and emergency dispatch in an urban, highly-populated county in the United States.
The workshop was organized to explore how a communications analyst working with social media
could support 911 call taking and dispatch, and the associated design requirements for awareness-
support tools that would enable the analyst in this work.
The workshop involved interviews with PSAP administrative (director, deputy director,
and operations manager), telecommunications (floor supervisors, call takers, and dispatchers), and
information technology staff (IT manager, CAD supervisor, and CAD technicians), as well as
approximately twenty hours of combined observation. During observation periods, the authors
more closely examined the workflows of 911 call takers, dispatchers, and resource managers, and
their interactions with Computer-Aided Dispatch (CAD), GIS, radio, and associated systems used
in emergency dispatch and response.
The workshop particularly focused on role-play activities in which six 911
telecommunicators volunteered to participate. Role playing involves a group of people who act out
roles in a constructed scene (Medler & Magerko, 2010; Svanaes & Seland, 2004). Widely used in
user-centered design, role playing allows researchers to observe plausible interactions among
participants, typically end-users and domain-experts, as they develop within hypothetical situations
168
(Medler & Magerko, 2010). Role play is especially useful for understanding activity in emergency
situations that are difficult to directly observe and inappropriate for research activities (Valkonen
& Liinasuo, 2010).
During the workshop, the authors specified the roles- caller, call taker, dispatcher, and
emergency responder, as well as citizen bystander and communications analyst- but the scenes were
left intentionally semi-scripted so that the telecommunicators could draw on their training and past
experiences to construct situations that were both realistic and critical from a design perspective.
Prompts included suggesting scenarios that the PSAP had encountered in the past, for example, an
active shooter in a local mall.
Qualitative analysis was performed on transcribed role-play sessions and post role-play
debriefing sessions. We performed content analysis using open coding strategies as suggested by
Strauss (A. L. Strauss, 1987) to develop a coding schema. Themes and concepts relating to
sensemaking breakdowns, interpretive frameworks, and sociotechnical infrastructures were
identified, discussed, and refined iteratively among researchers. For each role play, we analyzed
whether breakdowns in sensemaking occurred and utilized information gathered from post role-
play debriefing sessions to explore the occasions for each breakdown we observed.
Emergency Dispatch Simulations
This study reports findings from scenario-based simulations conducted at a large,
metropolitan PSAP which processes approximately 3000 calls a day. The role plays were organized
to provide the PSAP with prospective procedural and technological requirements for incorporating
social media in its emergency dispatch operations. Role playing involves a group of people who
act out roles in a constructed scenario (Medler & Magerko, 2010; Svanaes & Seland, 2004). Widely
169
used in human-centered design, role playing allows researchers to observe plausible interactions
among domain-experts and prospective end-users as they develop within hypothetical situations
(Medler & Magerko, 2010), especially emergency situations which are difficult to directly observe
and inappropriate for research activities (Valkonen & Liinasuo, 2010). The simulations involved
six telecommunicators performing in ensembles including a call taker, dispatcher (participants’
professional duties in the PSAP), and social media analyst, tasked with processing information
from synthetic 911 caller and social media datasets using a simulated Computer-Aided Dispatch
(CAD) system and analytic dashboard during two mock emergency scenarios.
During a prior workshop held at the PSAP in May 2018, researchers conducted activities
with six telecommunicators and 25 first responders (fire, medical, and police) to inform the design
of two scenarios and associated synthetic social media and 911 call datasets. First,
telecommunicators and first responders composed “golden tweets,” examples of social media posts
that would support situational awareness needs of first responders (blinded for review). Discussed
among the examples were golden tweets based on a mall shooting encountered by the PSAP the
year prior. These motivated the creation of the mall shooting dataset, consisting of 20 emergency-
related tweets, four mock 911 calls, and 1000 tweets collected for Charleston County on March
19th, 2018 from 4-6pm (as background noise).
Second, sets of 50 weather-related tweets were provided to first responders to sort into piles
of “actionable” and “non-actionable” information. The tweets were collected for Charleston County
on April 15th, 2018 when the PSAP experienced high call volumes during a severe storm that caused
flooding throughout the area. Using the sorted tweets and flood-related golden tweets, researchers
created the severe flooding dataset, consisting of three emergency-related tweets, 1000 “noise”
tweets collected on April 15th, and six mock 911 calls. Importantly, and inspired by our discussions
with telecommunicators and first responders, the mock tweets and 911 calls in both datasets provide
170
incomplete information (e.g. “what” without “where” information), and therefore require role play
participants to compare and integrate information across these reports to support the situational
awareness needs of first responders during each scenario.
Taking place during a follow-up workshop in August 2018, the simulations mimicked a
live emergency environment by placing the participating telecommunicators in the auxiliary
dispatch room of the PSAP in which the call taker, analyst, and dispatcher took up individual
workstations and interacted asynchronously via text-based communication using the simulated
CAD environment. The analyst was additionally provided with an analytic dashboard (Toepke,
2018), which included a map to examine geotagged tweets, tweet filtering and visualization using
search terms, and word cloud displaying trending hashtags that could be individually selected to
visualize associated tweets on the map and tweet list displays (Figure 5-2). A researcher with an
audio recorder was present at each workstation where talk-aloud methods were employed to collect
data from the call taker, analyst, and dispatcher as they processed calls or analyzed social media
data during each scenario (Jaspers, Steen, Van Den Bos, & Geenen, 2004). The participant in the
role of call taker communicated directly with the 911 caller (research confederate) and talked aloud
with a researcher while entering information in the simulated CAD.
Figure 5-2. Analytics dashboard interface (left) and scene from simulations with
telecommunicators and researchers at mock CAD workstations (right).
171
Using phases of the sensemaking cycle- framing, elaboration, questioning, and reframing-
as sensitizing concepts, researchers coded the transcribed audio recordings and textual data
participants entered in the simulated CAD environment (Strauss, 1987). The researchers discussed
and iteratively refined the coded data to address emergent aspects of the distributed sensemaking
process, for instance, the use of cues and search heuristics among telecommunicators when
searching and filtering tweets during each scenario.
Findings: Role Plays
We focus on one role-play scenario to illustrate sensemaking in emergency dispatch- to
include how social media data can contribute to this process- and examine the aspects of
sensemaking that were evoked throughout the role-play sessions. This role play witnessed 1) a 911
call break down when the caller did not provide requested information; 2) successive breakdowns
for dispatchers and responders who rely on this information; 3) sensemaking among multiple
officials who engage multiple data sources to address the resultant information gap; 4) the
communications analyst using social media data to provide the needed information that; as a result,
5) allowed the emergency response to resume.
We discuss each stage of the role play based around selected excerpts and giving special
attention to the relationships among breakdowns, interpretive frameworks, and infrastructures
through which we examine the sensemaking process.
172
“The caller is unresponsive at this time”
Two primary breakdowns emerge during the role play. The first develops immediately after the call
taker answers a 911 call:
Call Taker: 911, what is the address of your emergency?
Caller: Girl, you need to get here now and quit playin. You need to get them cops down
here now girl!
Call Taker: Ma’am, what is the address of the emergency?
Caller: Man, I know you can see me on the phone.
Call Taker: Ok ma’am, I need the address of the emergency.
Caller: Oh my god, I’m at North Romney.
Call Taker: Ok, do you know the address on North Romney?
Caller: 32B North Romney.
PSAPs set different targets for obtaining the location and chief complaint from callers. In the
scenario, the call taker asks four times before the caller describes the location with enough precision
for the call taker to dispatch a response. Without a specific location, call takers cannot process the
call for dispatch: “obviously we can’t respond if we don’t know where” (P4).
The role play continues as the caller reluctantly answers the call takers’ questions:
Call Taker: Thank you, tell me exactly what happened.
Caller: Girl, there’s about six people here ganging up on this little black bitch here. You
need to come get her…
Call Taker: What’s the phone number you are calling from?
Caller: Oh my god! I know you can see my phone number on that screen girl…
Call Taker: Does anyone have any weapons?
Caller: I... don’t know.
Call Taker: Ok, how many people are involved?
Caller: Man, there’s a bunch. There’s a bunch…
Call Taker: Do you know an approximate age range?
Caller: Man, they all from Bayside High School. Why you playin?...
Call Taker: By not answering these questions you may put you and responders at risk.
Please allow me to help you by answering these questions…[role play shifts]
The role play- introduced as “a common one for us down here” (P1)- quickly demonstrated what
the call takers often experience: uncooperative callers who fail to provide or simply do not know
173
the information requested. “When you are talking to a caller, 90% of our job they aren’t listening,
they’re not answering the questions” lamented one call taker during our follow-up discussion (P3).
Throughout the workshop, 911 telecommunicators described a data rich environment- this PSAP
processes over 3000 calls per day- and an information poor environment- callers often prove
uncooperative, unreliable, or simply ignorant of events occurring around them. “Probably better,”
replied the call taker when later asked about the quality of information reported on social media,
“because all I got is a screaming person on the line.”
As breakdowns occur when a contrast develops between the encountered and expected
course of an activity, 911 call taking breaks down when uncooperative callers interrupt the
standardized series of question and answer that allow call takers to obtain the information required
for dispatch to emergency responders. Call-taking sequences are structured around two
standardized protocols that allow call takers to efficiently question callers and sequentially enter
priority information into CAD. The “Six W’s”- Where, What, Weapons, Who, When, and Why-
provide call takers with a heuristic for questioning callers and entering only relevant information
for each call.
Second, ProQA, an expert system integrated into CAD, provides call takers with
emergency-specific question scripts and caller instructions, standardized text entry forms (i.e. call
notes). ProQA also assists call takers in determining the chief complaint code, the emergency
classification that determines the type and level of emergency, and, in turn, the police, fire, or EMS
resources that will be dispatched. Software such as ProQA, extend the “Six W’s” by walking call
takers through hundreds of standardized protocols for specific law enforcement, medical, and fire
situations.
These protocols tacitly and explicitly determine what information matters for call takers
during each 911 call: “Those are the most import things we need to know, everything else can,
174
honestly, be thrown in the trash, because that’s what we need to know” (P4). As only priority
information will be entered CAD- where, what, weapons, etc.- everything else a caller may say
during a call will be, in a sense, disposed of by the call taker. Protocols, as interpretive frameworks,
shape information gathering and filtering.
The role play next shifts to the dispatcher who, in reality, would be dispatching information
to responders at the same time as the call taker is on phone with the caller. The dispatcher receives
notice of the 911 call via CAD as soon as the call taker enters the call (after establishing the “where”
and “what”) and will dispatch information to emergency responders throughout the call by reading
the “call note” updates the call taker continuously enters into CAD while speaking with the caller.
Now joining the role play, the dispatcher begins using the (imaginary) radio to communicate with
police officers on patrol:
Dispatcher: [via radio] 32B N. Romney Street, 32B N. Romney Street, reference to an
active disturbance, multiple students, physical, units in route acknowledge
Responder: [via radio] 513 Newark copy… 513 Newark to dispatch, copy?
Dispatcher: Go ahead.
Responder: Does anyone have any weapons?
Dispatcher: Standby… unknown at this time. Call taker’s gathering additional
information.
Responder: 513 Newark copy.
Here the responder, a police officer in the scenario, easily enters the cooperative arrangement that
forms among caller, call taker, and dispatcher, as the officer requests information- the third “w” of
the Six W’s- in line with the same protocols for information gathering that are already guiding the
work of the call taker on the phone with the caller, and the dispatcher who waits for this information
to be entered in CAD. Against this common interpretive background, all the officials involved
recognize this absence of information- silence on the radio, a missing call note- as an information
gap that initiates distributed and cooperative sensemaking: each official in his or her own capacity
attempts to fill in information about on-scene weapons.
175
“Don’t forget to press the period”
The second breakdown emerges as the dispatcher waits for the call taker to forward call
notes regarding possible weapons at the scene of the fight. At this moment, however, the dispatcher
interrupts the role play to joke that the call takers’ notes might not have “dropped”:
P5 (Dispatcher): [Aside to call taker] Wait, don’t forget to press the period (laughter among
telecommunicators)... There is a flaw in our system right now...
P2 (Call Taker): [Explaining to authors] When we’re typing in stuff it’s not showing up-
dropping for them [dispatchers], so we have to hit a period and hit enter. So that it drops.
P5 (Dispatcher): And as dispatchers we recently learned that we can also do that, so you
will see like sixteen periods.
P3: I have learned if you just hit space and do it, it will drop, and it doesn’t put all the
annoying periods.
Communication between call takers and dispatchers takes place through “call notes” in CAD, a
shared text log among telecommunicators, in which call takers can enter priority information (Six
W’s) obtained from callers, and dispatchers can enter questions or updates obtained from
responders with whom they communicate by radio. When a telecommunicator enters or “drops”
information into call notes, that information will automatically update on other telecommunicators’
CAD interfaces. If, for example, the call taker cannot enter call information (i.e. if the caller is
uncooperative) or the call notes do not drop (i.e. if they do not automatically update on the
dispatchers’ CAD) then information will not reach the dispatcher and, in turn, the police officer
with whom she communicates via radio.
Though the telecommunicators laugh, the failure of call taker information to drop to
dispatchers, and vice versa, represents a serious problem: communication breaks down between
people requesting and offering help. The aside reveals the technical infrastructure that sustains
sensemaking in the PSAP: when call notes fail to drop, they occasion breakdowns in CAD as a
common information space (Bannon & Bødker, 1997; Wolbers & Boersma, 2013). As such, CAD
supports awareness of protocol-selected information dropped into the space among
176
telecommunicators and responders to whom this situational information is dispatched. As
infrastructure, telecommunicators interact with CAD to gather and share information to sustain call
taking and dispatch, respectively. At the same time, they re-construct and re-align this space with
each call as they draw on different infrastructures and data sources- radio, telephone, and GIS
systems- to search for and enter information to address information gaps that emerge in the common
information spaces of CAD.
“We out here in Bayside”
The role play shifts again. Two telecommunicators enter the role play as citizen bystanders
using social media. The first describes her likely actions: “Most of the tweets you are going to get
in that situation are... going to be other high school students. So, I would have tweeted a picture
with #BurkeHigh and #Fight, and let it go off from there” (P3). The second took a different
approach:
I did live video and commentary, Facebook Live... [begins commentary] “We out here in
Bayside, Bangin. Ya’ll we need the police, they’re always in here messin with us but they
ain’t here now.” And it’s just live of people fighting each other. (P4)
These social media posts imagine information that a future communications analyst could use to
address gaps emerging in 911 caller information.
However, when the communications analyst attempts to join the role play and contribute
to the sensemaking process currently underway, the telecommunicators pause to quickly coordinate
how this could work:
P3: [The comm. analyst] had better be putting in information that says…
P6: (comm. analyst): Wait, where are we? You just said, “no known weapon at this time.”
And [the Dispatcher] has relayed that to [the Responder] already?
P1: Correct, you’re relaying to [the dispatcher]. What you’ve done is opened up our call,
and you’re putting [call notes into CAD]... Your initials will be next to [the entered call
177
notes], and [the dispatcher] is reading the call as things are dropping and [the dispatcher]
is going to go back over the radio.
P5: (dispatcher): I have the call here [sketching CAD interface with her hands] and we have
a notification panel across the top, so you can either drop a note in [call notes] or send a
notification [which appears on the top of her screen].
P1: So, what did you put in the notes?
The telecommunicators position the new analyst vis-a-vis the same infrastructures that enable 911
call taking and dispatch. Just as a call taker now uses CAD to open a call and enter call notes while
questioning a 911 caller using protocols to gather and filter information, the analyst will
simultaneously open CAD and enter call notes while query social media data using the same
protocols. To know what matters in a time-critical situation then, the analyst will draw on standard
protocols such as the Six W’s:
P5: If you have multiple [data sources], [the comm. analyst is] going to have to sit there
and be able to sift through all that information, it’s not like he’s going to be able to see
everything at once. He’s going to be like alright, now I got that Facebook Live video, I’ve
got these Twitter posts coming up, and now somebody just posted something on Instagram.
P2: But that’s more when you just put notes in the call “Live feed showing handgun
present.”
P1: I would also assume that we would be just trying to get something particular if he is
involved… maybe he [comm. analyst] is not trying to get twenty questions, maybe he is
just trying to get one...
And at this moment in the role play, that one piece of information regards weapons. Such protocols,
as interpretive frameworks, make breakdowns apparent, revealing what information is missing
(e.g., are there weapons?), guide information gathering (e.g. are there indications of weapons on
social media?), and sharing (e.g. is information on weapons already available in CAD?) to enable
distributed, cooperative sensemaking among multiple officials working across multiple data
sources to synthesize information that address common situational awareness needs.
The role play then kicks off again, seamlessly:
Comm. Analyst: [via CAD] Per live feed, weapon on scene, handgun.
Dispatcher: [via radio] Dispatched units, updated information: We have a live feed
showing at this time that there are multiple subjects approximately [one handgun] can be
viewed at this time.
178
Comm. Analyst: [via CAD] Update: shots fired.
Dispatcher: [via radio] All units be advised shots fired 32 N. Romney, 32 N. Romney, shots
fired.
Responder: [via radio] 513 Newark copy. Dispatch, do we know who if anyone is injured?
Dispatcher: [via Radio] Unknown at this time. We started EMS en route, they will be
staging…
Caller: Whooooo, Whooooo, Whoooo, Oh my god! Somebody shot Kiki! Somebody shot
Kiki!
As the telecommunicators first explained, by using CAD to enter call notes during a developing
emergency the analyst, like a call taker, can contribute to the distributed sensemaking process
underway among the caller, call taker, dispatcher, and responder carrying out the emergency
response.
Findings: Simulation Role Plays
For each role play scenario, a synopsis describes the actions performed by the call taker,
dispatcher, and social media analyst, while annotations (in italics) associate these distributed
activities with sensemaking functions of framing, elaboration, questioning, and reframing. The
following discussion analyzes these sensemaking functions in detail by drawing on data collected
as call takers, dispatchers, and, especially, social media analysts, explained aloud their actions
during each scenario.
Scenario One: Mall Shooting
Scenario One begins when the Call Taker answers a 911 call reporting shots fired at the
Citadel Mall and, simultaneously, the Analyst is alerted to tweets describing a shooting in progress
(Table 5-2). The participating telecommunicators proceed to process two additional 911 calls and
179
multiple tweets reporting information on wounded victims, survivors sheltered in place throughout
the mall, and the description and whereabouts of the suspect(s).
Table 5-2. Four phases of scenario one.
CALL TAKER ANALYST DISPATCHER Someone dials 911: “I’m outside Citadel Mall and there are loud pops like someone is shooting here. I just ran out of target and am driving away…” The call taker enters an active assailant call [2] into CAD [Framing: Active Assailant], noting that there is no further information on potential suspects or injuries. [Elaboration: Information gap]
Alerted to tweets describing a “shooting,” searches and identifies three tweets and opens a call [1] for an active assailant on CAD [Framing: Active Assailant]: “several tweets reporting active shooter at the mall, one person reports near the food court and suspect is possible gone on arrival...one person reports 3 survivors sheltered in place in Journeys, negative injuries with this report.” [Elaboration: Survivors, Information gap]
Combines information entered by the Call Taker and Analyst for dispatch [Call 1,2]. Sends an alert tone for emergency units to respond to an active assailant incident in progress; begins to provide situational updates to first responders.
Searches map; identifies geotagged tweet reporting injuries and enters additional call notes [1] in CAD: “multiple shot near food court and shooter is gone on arrival, no descriptions or specific locations of victims or suspect.” [Elaboration: Victims, Information gap]
“...one party advising incident occurred near food court, suspect possibly gone on arrival, one party reporting three people sheltered in place at Journeys. Negative injuries reported at this time” [Call 1].
A second call comes in: “I’m in a back room in the store Belk. Someone is shooting and we are hiding here. There are three of us…” The call taker enters a new call [3]: “shots coming from inside the mall, poss. not IN Belk; negative on any suspect information; negative on any known injuries; weapon unknown.” [Framing: Active Assailant; Elaboration: Information Gap]
Uses the search term “Belk” to discover reports of shots fired near the department store. Comparing this information with call notes entered in CAD [3], the analyst concludes this information is redundant. [Questioning: Preserve frame (discard data)]
“...notification that there are still shots fired at this time, use caution. one caller advising they and two others are in Belk storage rooms sheltering in place… still no suspect info” [Call 3].
Another call [4]: “I am outside Belk in the parking lot… I think there are two white guys with guns… Lots of people are running through the parking lot…” The call taker makes further entries into CAD: “Parking lot outside of Belk; two white males, negative on clothing, unknown further; caller on the line, pending more suspect information.” [Elaboration: Suspects]
“Be advised, two white male suspects, armed, reported in the Belk parking lot” [Call 4]. In CAD, the dispatcher requests further information from the Call Taker on the line with the 911 caller: “...which parking lot of Belk?”
180
Framing the Emergency using Primary Cues
Alerted that “shooting” had been tweeted, the analyst adopts the word as a search term to
identify three tweets reporting an emergency:
#citadellMall #shooting Hiding in Journeys @NCPD 3 people here not injured @CCSO
There are people running and screaming in the mall. Someone is shooting
Shooting at citadel mall in Food Court. Shooter in just walked out
The Call Taker simultaneously answers a 911 call describing “loud pops, like someone is shooting
here...People were screaming and just running out of the store.” Based on primary information cues
(e.g. “Citadel Mall,” “shooter,” “loud pops”), the Call Taker and Analyst recognize the incident
type and location, and quickly enter calls 1-2 for an “active assailant” at Citadel Mall. By
establishing the primary information necessary to dispatch a response (i.e. where and what), the
call taker and analyst contribute to the emergency frame that will guide information seeking and
integration among the Call Taker, Analyst, and Dispatcher throughout the duration of the scenario.
Elaboration by Seeking Secondary Information
Next, the Analyst searches for secondary attribute information (e.g. weapons, who, when,
why) that will support situational awareness among first responders now en route to the scene.
Importantly, the emergency frame functions as a story, providing telecommunicators with a likely
order of events taking place at Citadel Mall, as well as a set of corresponding information
requirements:
For the initial response I want to know who the suspect is and what they look like and
where they go… several people have reported that they’ve ran away so did they run out of
the food court, did they run out of where? Did they get into a car? Everything about the
suspect.
Once the suspect issue has been dealt with by the police department, then… they can find
out where the victims are… From there we would go into where people are sheltered in
181
place. So suspect information first, then where the victims are, and then the sheltered in
place. (Analyst, Scenario 1)
The emergency frame, in this case an active assailant incident at Citadel Mall, prioritizes situational
awareness (i.e. secondary information) requirements to guide information seeking between the call
taker and analyst as they search for new information and, ostensibly, dispatch information first
responders need when they need it.
In the scenario, however, this never happens. The Analyst adopts ineffective search terms
and fails to discover most of the suspect information available in the tweet dataset during the
scenario, notably:
I’m locked in the women’s room. There’s a tall skinny white guy in blue jeans and a white
tshirt.
man with a gun just got into a red SUV and turned right on Orleans Rd
shirtless man with assault rifle running to car outside the mall [picture of shirtless man next
to a red SUV]
During both scenarios, telecommunicators rely on search heuristics employed as 911 call takers,
but which prove ineffective when adapted to filtering social media data (i.e. poor recall). Explaining
that he has “seen a lot of people using the business name versus an address or road name,” the
analyst uses the department store name “Belk” referred to by an earlier 911 caller (Call 3). Relying
on cues provided by 911 callers, analysts adopted similar heuristics, typically using the names of
businesses or other geographic landmarks, to search generally for emergency-related information.
Consequently, telecommunicators did not select search terms associated targeting specific,
priority information. Instead, analysts provided information for dispatch as discovered, such as
when the analyst provided information on survivors sheltered in place (Call 1) early in the response
when suspect and victim information remained priorities. By using search terms explicitly provided
by 911 callers or alerts provided by the analytic dashboard, the telecommunicator-as-social media
182
analyst reactively filtered social media data in ways that often proved ineffective at identifying and
triaging priority information.
Elaboration by Identifying Gaps in Situational Awareness
Whenever entering new information into CAD, the Call Taker and Analyst also described
what information remained unknown: “negative injuries with this report” (Call 1), “unknown any
injuries… no suspect information” (Call 2), “negative on any suspect information; negative on any
known injuries; weapon unknown” (Call 3), “negative on clothing, unknown further” (Call 4). By
identifying information gaps in CAD, the Call Taker and Analyst support shared awareness among
telecommunicators and first responders during the response:
When we got the additional tweet of multiple people shot, I went back to the original [call]
notes I had [for Call 1] to report multiple people shot near the food court... the shooter is
gone, and... that there are no descriptions or specific locations because some of these
officers are going to ask for that. I know they are going to ask for that so I go ahead and
tell them there is nothing more specific. (Analyst, Scenario 1)
Information gaps can be determined with respect to secondary information associated with the
emergency frame: an active assailant at Citadel Mall. In Scenario One, missing information
included descriptions of weapons, the shooter, as well as potential victims and survivors sheltering
in place. As previously described, however, the analysts’ inflexible search heuristics precluded
targeted information seeking to address information gaps articulated among the telecommunicators.
183
Questioning and Discarding Redundant Information
In phase three of the scenario, the Call Taker enters a call (3) from a survivor sheltering in
place in Belk, a department store in Citadel Mall. The Analyst subsequently uses the search term
“Belk” and identifies two tweets:
loud pops near Belk in the mall. Sounds like a gun people running screaming
In Belk 🙏🙏🙏 #help #pray #charleston
The Analyst then compares this information to existing information entered in CAD and already
dispatched to first responders:
I searched for “Belk.” I had one [tweet] pop up basically reporting what sounds like pops,
or a gun going off… There is no additional information I can pull out of that. They already
know there is a shooting at the mall near Belk, they already know people are running
around and screaming. It’s an active shooter incident. So, there is nothing additional I can
pull from here to put into there… It’s redundant information because everything that is
here is already here [in CAD]. (Analyst, Scenario 1)
Although the tweets provide information related to the emergency, the Analyst does not consider
them relevant in the situation given the information already available and the evolving information
requirements of the response. From a sensemaking standpoint, the phase illustrates the Analyst
questioning encountered information to gauge its quality with respect to information characterizing
the existing frame. Concluding the former corroborates the frame without providing any new
information for its elaboration, the Analyst discards the data.
Scenario Two: Severe Flooding
Scenario two involves roadway flooding in different locations causing traffic hazards and
the need for two water rescues. Table 5-3 provides a synopsis of the role play in three phases. As
in the mall shooting scenario, the call taker, analyst, and dispatcher encounter incomplete
184
information across each information source: 911 callers and social media users provide only some
of the information telecommunicators and first responders seek during the three emergencies
included in the scenario.
Table 5-3. Three phases of scenario two.
CALL TAKER ANALYST DISPATCHER Enters a call [1] into CAD when someone reports roadway flooding near the Dollar General on Camp Road. [Framing: Traffic Hazard]
Checking tweets appearing on the map, then searching for “Dollar General” and “flooding,” discovers tweet reporting a trapped person trapped at the intersection of Dills Bluff and Camp Road. Determines this a distinct incident and enters a call [2] into CAD. [Questioning: Inconsistent location; Reframing: Water Rescue]
Uses radio to dispatch a police unit to Camp Road [1], and then police, fire, and medical units to the Dills Bluff incident [2], noting: “units dispatched to both sides of the roadway to set up cones and Traffic and Transportation notified [of lane closures].”
Also discovered when searching for “flooding,” a tweet shows an image of flooding in the parking lot of Harris Teeter. Enters a call [3] into CAD. [Framing: Traffic Hazard]
Dispatches a police unit to the Traffic Hazard near Harris Teeter [3] and asks the Analyst to clarify the depth of the water in the tweeted picture: “what is a 4-5 drop zone?”
Another call arrives reporting more roadway flooding, this time near Pier One Imports on Sam Rittenberg Road. Enters another call [4] for a Traffic Hazard into CAD. [Framing: Traffic Hazard]
Searching “Pier One,” finds tweet reading “So much water at Lake Francis near CVS. Rd. impassable. ONE PERSON STUCK IN CAR.” Determining that the CVS is in the same parking lot as the Harris Teeter, enters call notes [Call 3] requesting a Water Rescue. [Questioning: Inconsistent location; Elaboration: Water Rescue]
Dispatches police to Traffic Hazard on Sam Rittenberg and uses CAD to ask the Call Taker if anyone is in the vehicles. “Caller did not advise,” answers the Call taker. Dispatches units to a water rescue. In call notes [for Call 3] the dispatcher writes, “Copy, duplicating call for emergency Fire Rescue (EFR). Confirming information with police department on the scene.”
Discovering Inconsistencies among Information from Social Media and Emergency Callers
As in scenario one, the analyst did not readily discover information during each emergency,
to include the discovery of inconsistencies among information that might lead to a reframing of
awareness. In phase two, for instance, the Analyst enters a call for a traffic hazard on Harbor View
Boulevard after searching “flooding” and discovering a tweet reporting flooding on Harbor View
near Harris Teeter, a supermarket. What the analyst did not discover, however, was a tweet
requesting a water rescue in the same location:
185
So much water at Lake Francis near CVS. Rd. impassable. ONE PERSON STUCK IN
CAR.
Lake Francis Drive intersects with Harbor View Boulevard immediately across from the shopping
center parking lot in which both Harris Teeter and CVS, a pharmacy, are located. The analyst later
discovers this tweet by accident when searching for “Pier One” after the store name was mentioned
by a 911 caller reporting flooding eight miles away on Sam Rittenberg Boulevard- an unrelated
incident. That the search results for “Pier One” returned a tweet reporting “One person stuck in a
car” was purely serendipitous.
As the example illustrates, the types of cues provided by 911 callers and entered into CAD
by call takers condition analysts’ discovery of emergency-related information. The previous
example also illustrates the difficulty of discovering an emergency when no one calls 911; the
initial call entered into CAD for a traffic hazard on Harbor View near Harris Teeter was entered
earlier by the analyst after discovering a picture of the flooding on Twitter. That the analyst was
not initially aware of the trapped person at the Harborview and Lake Francis intersection
demonstrates, again, the difficulty of discovering unknown information about an unknown
emergency.
Evaluating Inconsistencies among Information from Social Media and Emergency Callers
Searching and filtering tweets, however, constitutes only the first step of integrating
information from social media within 911 dispatch operations. Next analysts must compare newly-
obtained information from social media displayed on the dashboard interface with existing,
integrated information from 911 callers and social media displayed on CAD and shared among the
call taker, dispatcher and analyst. At this stage inconsistencies appear when analysts cannot not
186
determine if a discovered tweet(s) reports a new emergency not yet entered into CAD, new
situational awareness information for a reported emergency, or redundant information about an
already-reported emergency.
Consequently, the ability of analysts to evaluate inconsistencies among information
displayed on the analytic dashboard, and between information displayed on the dashboard and
CAD, becomes critical to effectively integrating information from social media and 911 callers for
emergency dispatch. In phase one, for example, the analyst must determine if the trapped person at
the intersection of Camp Road and Dills Bluff is related to the flooding earlier reported on Camp
Road (near the Dollar General) by a 911 caller. Comparing the available location information for
each incident, the analyst determines them to be separate and therefore enters a new call into CAD
for a water rescue, rather than adding additional information to the existing call for the traffic
hazard.
To do this, however, the analyst uses primary information cues in the tweets, words that
indicate the incident type and location of the emergency. When asked to consider a likely
counterfactual, if the tweet had provided less detailed location information by omitting reference
to Dills Bluff, the Analyst describes how telecommunicators evaluate and compare inconsistent
information from multiple emergency reports:
If this Dills Bluff thing wasn’t here, and it just said lots of flooding on Camp Road, my
car’s flooded, I need a boat, [then] flooding on Camp Road is related to the Dollar General
on Camp Road where there is a lot of flooding reported. I would add it to the same call
then. However, because there is this specific location, I would probably create an entirely
separate call. (Analyst, Scenario One)
Lacking the primary cue “Dills Bluff,” and therefore presented with less exact location information,
the analyst would infer that the traffic hazard and water rescue are colocated- both on Camp Road
near Dollar General- and refer to the same roadway flooding. However, when presented with more
187
detailed location information, the analyst concludes that the incidents are distinct, occurring at
different locations along Camp Road.
The analyst is working with protocols for dispatching a response that require classifying
an incident type- in this case a water rescue- and an exact location. If the cue “Dills Bluff” was
omitted by the social media user or not presented to the analyst, only the Camp Road at Dollar
General location would meet this latter criterion. From a sensemaking perspective, the example
illustrates how the analyst relies on cues in data to either elaborate an existing frame (Camp Road
at Dollar General) or construct a new frame using cues sufficient to classify an exact location (Camp
Road at Dills Bluff). Importantly, the analyst relies on the bottom-up presentation of cues available
in the data and, at the same time, the top-down recognition of sensemaking frames from the
repertoire of locations available for dispatch in Charleston County.
Similarly, in phase three, when the analyst serendipitously discovers the tweet requesting
a water rescue on Lake Francis Boulevard near CVS, he must still evaluate the tweet in comparison
to inconsistent information entered into CAD for dispatch:
So when I was looking for Pier One there weren’t any geotagged tweets from the area [on
the map] but it looks like there are additional ones that popped up [in the tweet search]. I
found one tweet here, “water at Lake Francis near CVS, road impassable, one person stuck
in the car.” So now I would identify where this is, looking up Lake Francis, looking up
some of the CVSs trying to figure out where this was. (Analyst, Scenario 2)
The report of someone trapped in their car near Lake Francis near CVS is not immediately
intelligible as the same location as the traffic hazard on Harbor View near Harris Teeter. Both
roadways are long thoroughfares, and there are multiple chain supermarkets and pharmacies with
the same name in Charleston.
To compare this new information with that already entered into CAD, the analyst again
looks to primary information cues in the tweet that allow telecommunicators to classify information
for emergency dispatch:
188
I can use the CAD system to determine where is Harris Teeter, where is Lake Francis, and
where is the CVS near Lake Francis.... I would just look for Lake Francis Drive and see if
I could find a CVS near there, and then compare this report to the CAD to see if there was
a call already entered and, if not, then I would go ahead and enter one for water rescue. So
they wouldn’t be getting much information but they would be getting enough for a location
and a basic response for one person stuck in a car due to possible flooding. (Analyst,
Scenario 2)
Using the names of roadways and stores as primary cues, the analyst uses mapping applications
(typically integrated in CAD) to classify the location of the emergency and determine if an
emergency has already been reported in the same location. In this situation, the analyst determines
that the newly-reported water rescue, and the already-dispatched traffic hazard are colocated: “if
this is the same location...the person stuck in the water in the car is due to the flooding, basically”
(Analyst, Scenario 2).
Discussion
As emergency dispatch centers such as PSAPs already coordinate the flow of information
between citizens and first responders, incorporating social media requires 911 call takers,
dispatchers, and prospective social media analysts to make sense of information gathered from
social media using information gathered from emergency callers, and vice versa. In this regard,
incorporating social media analytics in emergency dispatch continues the transition of PSAPs from
reactive call centers to proactive data analytics and coordination hubs using Next-Generation 911
infrastructure to collect and process heterogenous data from physical and social sensors to support
the situational awareness needs of first responders. In both free-form and simulation role plays, 911
call takers, dispatchers, and social media analysts utilized a simulated CAD environment and
analytic dashboard to process and integrate information from social media and 911 callers during
mock emergency scenarios. The role plays suggest procedural and technical requirements for
189
enabling distributed sensemaking processes among multiple actors, technologies, and data sources
now characterizing next-generation emergency dispatch work.
Overall, the free-form and simulation role plays reveal how the distinct domain ontology
organizing emergency dispatch work coordinates information analysis, seeking, and integration
among multiple actors engaging multiple technologies and data sources. Findings contribute to
theory surrounding social media, sensemaking, and situational awareness by showing that social
media content does not, ipso facto, enhance situational awareness in emergency response unless
coordinated within the distributed sensemaking processes of emergency responders (Baber &
McMaster, 2016; Kropczynski et al., 2018; Zade et al., 2018). Social media cannot be simply
“pumped in” to officials but must be coordinated within existing workflows in which it provides
incomplete information only in relation to other incomplete information sources. Situational
awareness is, therefore, the achievement of domain-dependent processes that coordinate the
integration of information across multiple, incomplete, but complementary data sources to meet
unfolding information requirements during an emergency. To meet these requirements, the
simulated role plays suggest that social media analysts will require new protocols and software to
help analysts proactively address evolving information needs during an emergency.
Emergency Dispatch as Distributed Sensemaking
To understand these processes, we focus on three interconnected aspects of sensemaking:
breakdown, interpretive frameworks, and sociotechnical infrastructure. From the perspective of
each aspect we present our findings regarding the integration of communications analysts working
with social media data in Public-Safety Answering Points (PSAPs), and implications for research
190
examining social media data as a source of situational information that can contribute to emergency
response.
Local Breakdowns Invite Distributed Sensemaking
Breakdowns occur when 911 call takers and dispatchers encounter missing information
that interrupts the course of their work. When a 911 caller proves unresponsive and the scripted
question and answer dialogue breaks down, or when entered call notes fail to “drop,” call takers
and dispatchers encounter information gaps that occasion breakdowns in PSAP workflows. As
PSAPs rely almost exclusively on 911 calls for situational information, information gaps resulting
from breakdowns in call taking occasion breakdowns for dispatchers, and, in turn, emergency
responders. The tendency for successive local breakdowns in activity identify emergency dispatch
and response as inherently distributed and mutually-dependent activities. That is, they involve
cooperation and require coordination [33].
However, as the role play demonstrates, local breakdowns can be overcome when
distributed officials become aware of the information gap and can draw on alternative information
sources to share insights that allow response activities to resume. During the role play, the
communications analyst using social media data was able to provide unique information about on-
scene weapons that, in turn, allowed the dispatcher to resume her information updates to the
emergency responder. Here the 911 telecommunicators turned to social media for select situational
information that was unavailable when relying on information provided by the 911 caller alone.
Critically, the communications analyst was able to draw on interpretive frameworks and
sociotechnical infrastructures that conditioned his awareness of the information gap, and guided
his (imaginary) analysis of social media data and selection of situational updates entered into CAD.
191
More generally, we find emergency dispatch to involve multiple, incomplete information
sources (i.e. 911 callers) that, individually, often fail to support the information needs of emergency
dispatchers and responders. This study suggests, then, that in a data-rich environment the
availability of situational information on social media is distinct from the utility of that information.
As emergency responders rely on multiple data sources, the extent to which social media data can
enhance situational awareness depends on the information content obtainable on social media and
the extent to which social media data can be integrated into distributed sensemaking processes that
coordinate the synthesis of unique information across these data sources.
Our findings therefore contribute to theory surrounding social media and situational
awareness by showing that social media content does not, ipso facto, enhance situational awareness
in emergency response unless coordinated within the distributed sensemaking processes of
emergency responders. Social media cannot be simply “pumped in” to officials but must be
coordinated within existing workflows in which it provides incomplete information only in relation
to other incomplete information sources. Situational awareness is, therefore, the achievement of
domain-dependent processes that coordinate the synthesis of information across multiple,
incomplete, but complementary data sources to meet unfolding information requirements during
an emergency. As a result, opportunities to use social media data to enhance situational awareness
require aligning the “inputs” of social media with coordination mechanisms that organize this
distributed sensemaking processes.
Protocol as Interpretive Framework
The role play demonstrated how protocols such as the Six W’s and ProQA serve as
interpretive frameworks during sensemaking processes: showing officials what information matters
192
during an emergency and coordinating information gathering, filtering, and sharing among multiple
officials working with multiple data sources. Such protocols are domain-dependent and are critical
in enabling the cooperative sensemaking processes that support emergency responders’ situational
awareness.
Studies that seek to understand what types of situational information is available on social
media during an emergency using qualitative coding [25, 31, 39], and those that use qualitatively-
coded datasets to develop machine learning classifiers to filter situational information posted on
social media [15, 18], stand to be improved by adopting the criteria for situational information
explicit or implicit to the domain-specific protocols of emergency response practitioners.
While machine learning approaches have attempted to identify tweets related to an event
lacking common keywords or hashtags [18] or explicit location information [20], understanding
domain-specific protocols can help refine classifiers to more accurately filter social media data and
identify information that supports emergency responders’ sensemaking and situational awareness.
As described by telecommunicators during the workshop, PSAPs typically gather
information pertaining to the Six W’s, often using the scripted questions provided by ProQA that
to obtain information required by emergency responders. Tools supporting future communications
analysts will be similarly required to assist analysts by, for example, pre-filtering and visualizing
social media data in ways that align with domain-dependent information requirements. Studies that
evoke the tacit and explicit protocols of these domains can provide criteria for qualitatively coding
social media datasets that, in turn, can inform the development of automated classification methods.
193
Sensemaking Infrastructures
Lastly, sensemaking infrastructures consist of human and artifactual (e.g. information
systems) resources for action [34] that sustain inquiry during sensemaking processes. In the context
of the PSAP, we find three intertwined infrastructures: the protocols shared among call takers,
dispatchers, and responders, the distributed communication technologies and data sources these
officials engage, and the Computer-Aided Dispatch (CAD) system that provides a common
information space through which officials can access and share information. Officials draw on and
interact with these infrastructures to reveal information gaps associated with data sources (e.g. 911
callers) and cooperatively gather and share information to address these gaps.
As the role plays suggest these infrastructures- as resources for action- are mutually
constituted: CAD, as a common information space, requires officials to enter and share information
obtained from different communications channels and data sources (telephone, radio, social media).
However, the information officials enter and share is that which addresses information gaps
recognized on CAD. As described, protocols shared among officials- as interpretive frameworks-
serve as infrastructures and coordination mechanisms that facilitate sensemaking processes by
allowing officials to recognize and address common information gaps appearing on CAD. We find,
then, that prospective communications analysts should be prior telecommunicators or receive the
same training in protocols guiding 911 call taking and dispatch. Furthermore, communications
analysts will require the same CAD workstation as telecommunicators, in addition to social media-
specific tools that can collect, process, and visualize social media data, so that they can integrate
protocol-selected information obtained from social media data into the common information spaces
that enable sensemaking processes in the PSAP.
194
Design Requirements for Incorporating Analysts in Emergency Dispatch
In each scenario of the simulation role plays, the call taker, analyst, and dispatcher
cooperate to understand an unfolding emergency as they respectively answered 911 calls, searched
for tweets, and dispatched emergency units. In so doing they demonstrate a sensemaking process
in which functions of the sensemaking cycle- framing, elaboration, questioning, and reframing- are
distributed among multiple people using multiple technologies to analyze multiple data sources.
However, the role plays also suggest that social media analysts will require new protocols and
technological assistance enabling proactive information seeking and integration, visualizations that
highlight inconsistencies among social media and 911 caller data, and guidelines for evaluating
information from social media during emergency dispatch situations.
Analyzing Information from Social Media and 911 Callers: Framing the Emergency
The simulations demonstrate how dispatchers make sense of social media data in sequences
of sensemaking activities, or workflows, performed among members of the dispatch team to
process information that supports the evolving situational awareness needs of emergency
responders. Drawing on Klein et al. (2006, 2007), this study observes four sensemaking activities
constituting sensemaking workflows: framing, elaborating, questioning, and reframing. For each
sensemaking activity, this study points to design implications for social media analytics that can
help analysts make sense of social media data and protocols for coordinating analysts’ sensemaking
activities in workflows performed among emergency dispatch teams (Table 5-3) (Kropczynski et
al., 2018; Zade et al., 2018).
The first activity, framing, sees dispatchers-turned-analysts make sense of social media
data as they would 911 calls: associating data cues with a frame(s) explaining the situation. To
195
analyze cues in social media data, analysts draw on the repertoire of frames commonly referred to
as the “6Ws”: Where, What, Weapons, Who, When, and Why (Kropczynski et al., 2018). The 6Ws
represent a domain ontology organizing emergency dispatch work in which determinant codes
representing classes of emergencies (i.e. what) are defined by attributes (i.e. where, weapons, who,
when, why) with a set of possible values (e.g. street address in jurisdiction). Thus, analysts frame
cues such as “loud pops,” “shooting,” and “Citadel Mall” with a determinant code indicating the
class of emergency (e.g. 136E-1: Active Assailant) and a street address (e.g. 2070 Sam Rittenberg
Blvd, Charleston, SC 29407). Thus, for analysts, effectively analyzing social media requires social
media analytics that can filter domain-specific information whose relevance is defined by the
presence of cues enabling analysts to classify the 6Ws associated with an emergency (Kropczynski
et al., 2018; Zade et al., 2018).
Table 5-3. Design requirements supporting sensemaking by social media analysts.
(Re)Framing
Cue Filtering Identification and visualization of 6W cues in social media posts
Domain Ontology Incident determinant codes and information requirements (6Ws)
COP Activity awareness of calls and responding units
Elaborating Assisted Search 6W-based classifiers to discover targeted information
IS Protocols SOPs to proactively address information gaps
Questioning Alerts Notifications to initiate sensemaking
6W Visualizations Visualizations contrasting 6W cues to reveal inconsistencies
Framing practically involves entering a “call” in CAD whose primary information includes
the “where” and “what” associated with an emergency, the minimum information required to
dispatch emergency units. Highlighting the role artifacts as frames (McMaster et al., 2012), and
paired with the 6Ws domain ontology, calls displayed in CAD create a Common Operating Picture
(COP) for a dispatch team, providing awareness of available information, information gaps, and
the progress of emergency response operations. Together, cue filtering, the 6Ws domain ontology,
and the function of CAD as a COP, enable analysts to create frames which, in turn, condition
196
subsequent opportunities for distributed sensemaking activities of elaborating, questioning, and
reframing.
Integrating Information from Social Media and 911: Elaborating the Emergency Frame
When a call is entered in CAD, the second phase of sensemaking proceeds as call takers
and analysts seek secondary information that will elaborate the existing frame by providing
additional details. As a frame, call information in CAD orients analysts’ information seeking by
articulating information requirements associated with gaps in “weapons,” “who,” “when,” and
“why” information prioritized according to the evolving needs of emergency responders. The
simulations demonstrate how 911 calls entered in CAD provide analysts with resources for
searching social media data and coordinating information seeking between call takers and analysts.
However, the simulations also suggest analysts require assisted search features and information
seeking protocols.
The CAD call, as a frame, provides the dispatch team with a narrative of events happening
during an emergency which analysts can, theoretically, use to search for priority information. In
scenario one, for instance, the analyst knows that responders require immediate details about
suspect(s) responsible for the mall shooting rather than survivors sheltering in place. However,
while analysts use frames as stories to understand how information requirements associated with
chronological events will likely evolve during an emergency, the simulations suggest analysts must
also use frames as scripts to coordinate proactive information seeking among dispatch teams. Such
protocols would direct analysts to search social media data for priority information addressing the
information gaps dispatchers describe in CAD whenever entering new information. Moreover, such
197
protocols would encourage analysts to proactively search for missing information rather than
reactively search for information when prompted by 911 callers.
Proactive information seeking, however, requires the ability to search and discover priority
information. During both scenarios, analysts’ search heuristics proved ineffective when filtering
large social media datasets. To select search terms, analysts adapted techniques developed as call
takers and searched for place names (e.g. “Camp Road,” “Belk”) mentioned by 911 callers. These
queries resulted in poor recall of relevant information. Consequently, social media analytics for
emergency dispatch will likely require assisted search capabilities leveraging, for instance,
keyword-based classifiers that would allow an analyst working with a single cue, such as the
department store name “Belk,” to filter social media data associated with all toponyms in that
geographic location. Similar approaches could be leveraged for each of the 6W’s, allowing, for
example, an analyst using the cue “shot people” to filter social media data using keywords and n-
grams associated with potential victims, i.e. “who.” Future efforts to implement assisted search
features in social media analytics dashboards can draw on the extensive research surrounding
automated classification techniques for crisis-related information (Imran et al., 2015), but must
tailor these techniques to the domain ontology organizing information processing in emergency
dispatch work (Kropczynski et al., 2018).
Evaluating Inconsistent Information: Questioning and Reframing the Emergency
When dispatchers discover inconsistencies, they must evaluate if the discovered
information is novel, supplementary, or redundant in relation to information characterizing the
existing frame. If determined novel, call takers or analysts can reframe the emergency by entering
a new call in CAD or replace an existing call. Reframing produces a new frame for the dispatch
198
team that, in turn, re-starts the distributed sensemaking cycle. For analysts, the emergency frame
comes into question when noticing inconsistencies between newly-discovered information on
social media and call information in CAD characterizing the existing frame. Analysts notice
inconsistencies when comparing among information discovered on social media (e.g. do tweets
refer to different locations?) and between information from social media and information in CAD
(e.g. do tweets refer to the same location as 911 callers?).
During the simulations, however, analysts recognize inconsistencies rarely and, in the case
of the “Pier One” example, often inadvertently. Instead, analysts face a common problem in
sensemaking: you don’t know what you don’t know. Consequently, analysts will likely require
automated alerts to discover information about potential emergencies not yet reported by 911
callers. In the mall shooting scenario, the alert notifying the analyst that the word “shooting” was
tweeted provides an example. Important questions regarding effective keywords, frequency
thresholds, and other measures for triggering alerts, however, extend beyond the scope of this study.
Furthermore, analysts will likely require information visualizations that not only highlight relevant
cues in social media data but draw attention to inconsistencies among (e.g. comparing all available
location cues) and between (e.g. comparing location cues against incident cues) 6W information
discovered on social media, and, furthermore, between 6W information discovered on social media
and call information in CAD. The analytics dashboard employed during the simulations visualizes
social media data in typical ways: plotted on a map, if geocoded, or filtered by search criteria in a
social media stream. Understanding the domain ontology of emergency dispatch work, and how
dispatchers use cues to frame social media data, can inform domain-specific visualizations that may
help analysts detect inconsistencies challenging existing frames.
199
Conclusion
Through role-play activities with 911 call takers and dispatchers we illustrate how future
communications analysts working with social media data can contribute to processes of
sensemaking in PSAPs. By attending to aspects of sensemaking- breakdowns, interpretive
frameworks, and sociotechnical infrastructures- this study provides insight into the coordination of
sensemaking processes that can enhance situational awareness in near-future environments of
distributed emergency reporting, dispatch, and response. The simulations, in turn, suggest that
social media analysts cannot adopt the reactive posture of call takers waiting on incoming 911 calls
but must, instead, proactively seek and integrate priority information during emergencies. This
proactive role likely requires automated event detection and alerts, assisted search capabilities to
increase the recall and precision of analysts’ search queries, and new protocols that coordinate
priority information seeking to address information gaps that evolve during an emergency response.
Important limitations attend the simulated character of emergency dispatch work examined
in this study. Despite involving professional dispatchers in the design of the simulations, including
the construction of the social media datasets, real-world conditions will of course vary significantly.
Issues such as the need to verify information reported on social media were not fully addressed in
this study although they remain an import consideration in emergency responders uses of social
media (Hiltz & Plotnick, 2016; Reuter et al., 2016). The nature of the simulations, moreover,
recommends that the design requirements suggested by the findings require further investigation
and empirical evaluation. For instance, the design of alerts, as Boersma et al. (2016) observed,
remains a difficult balance between recall and precision: too many undermine the value and
effectiveness of user notifications while too few risk missing critical information. Moreover, the
utility of alerts will necessarily depend on the information behaviors of social media users in a
geographic community and the envisioned use cases for emergency dispatch and response.
200
Furthermore, additional research is required to design and evaluate visualizations supporting
sensemaking activities such as questioning. Again while prior research must inform initial designs
(Ntuen, Park, & Gwang-Myung, 2010; Stasko, Görg, & Liu, 2008), the domain-specific nature of
the design will require iterative ideation and evaluation engaging dispatchers as domain experts
and prospective end users.
Overall, the simulations conducted in this study suggest that social media analysts cannot
adopt the reactive posture of call takers waiting on incoming 911 calls but must, instead, proactively
seek and integrate missing information with information already available from 911 callers. This
proactive role will likely require effective filtering and automated alerts to notify the analyst of new
and possibly inconsistent information, assisted search functions to increase the recall and precision
of analysts’ search queries, and information seeking protocols to address information gaps that
evolve during an emergency response. Furthermore, analysts stand to benefit from custom
visualizations that support top-down analysis by organizing and contrasting 6W information to
assist sensemaking activities of framing, elaborating, questioning, and reframing.
201
Chapter 6
Conclusion
Community emergency response consists of cooperative and coordinated activities
performed by professional and non-professional actors in a geographic community immediately
before, during, and after an emergency to suppress and mitigate human and environmental impacts.
To understand the incorporation of social media in this context, Phase I asked, “What objectives,
cooperative tasks, and coordinative practices characterize social media use in community
emergency response?” In contrast to prior studies, Phase I recontextualized social media
distribution and monitoring in the activity context of emergency response and revealed the
coordinative role of Public-Safety Answering Points (PSAPs) in detecting, verifying, integrating,
and dispatching information gathered from social data sources— social media users and 911
callers— to emergency responders. The interviews with emergency responders revealed two
requirements for incorporating social media in community emergency response which were
addressed in Phases II and III, respectively: methods collect hyperlocal social media data and new
workflows to incorporate social media analysts and social media analysis tools in PSAPs.
Phase II took up the first problem: “How can emergency responders identify hyperlocal
social media users and collect hyperlocal social media data?” Introducing the dual methods of
Social Triangulation and network filtering, Phase II demonstrated opportunities to collect more and
different social media data than existing location and keyword filtering data collection methods.
Furthermore, Social Triangulation identifies hyperlocal social media users and information sources,
providing opportunities to understand who uses social media in a community and how to reach
them with official messages. The latter informs emergency communications planning by
identifying “filter bubbles” among citizens using social media in a community, “community
202
influencers” positioned to reach these users by re-distributing official information, and citizens’
organizations whose followers might play useful roles relaying information across uneven
community networks.
Lastly, Phase III moved inside the walls of the PSAP to answer the final research question:
“How can emergency responders integrate information from social media and 911 callers in
emergency dispatch operations?” To do so, the studies of Phase III examined existing and
prospective workflows among 911 call takers, dispatchers, and social media analysts. The role
plays and simulations performed among dispatchers suggested requirements for new procedures
and software to help analysts make sense of information posted by social media users for integration
with information obtained from 911 callers. Phase III also demonstrated distributed sensemaking
processes enacted within PSAPs during an emergency, and how the introduction of social media
analysts stand to reshape these processes. Addressing the workflows observed among emergency
call takers, dispatchers, and analysts as distributed sensemaking processes, Phase III identified
elemental workflows— framing, elaborating, identifying redundancy, and reframing— required for
emergency dispatch work integrating information gathered from multiple social data sources. These
workflows outline future activities for the design of procedures and technologies that can enable
Next-Generation 911 (NG911) dispatch operations.
The following conclusions consider, first, practical implications for introducing social
media analysts and analytics in community emergency response broadly, and for dispatchers
serving in PSAPs particularly. These implications must be considered not only regarding
prospective uses of social media, but the impending changes to emergency dispatch work that will
accompany the introduction of NG911 infrastructure. Second, the theoretical implications of the
studies emphasize the need to move Crisis Informatics research from the perspectives of
“situational awareness” to a perspective of sensemaking that recognizes the contextual and
203
processual activities of emergency response work within which information gathered from social
media becomes actionable and non-actionable. A sensemaking perspective also shifts research from
how people use social media to research examining how people use multiple information sources
during emergencies. Lastly, the methodological implications of this dissertation recommend design
approaches in Crisis Informatics that move from low to high-fidelity enactments revealing, first,
activities of information integration and distributed sensemaking and, second, sociotechnical
requirements to enable these activities.
Implications for Community Emergency Response
Introducing social media monitoring in PSAPs creates new opportunities and challenges
for first responders and, especially, telecommunicators-turned-analysts responsible for discovering
and integrating information posted on social media in transformed emergency dispatch workflows.
While first responders such as fire fighters, paramedics, and police officers may not notice
immediate changes in how information is dispatched, incorporating social media monitoring in
PSAPs stands to change the nature of information available before, during, and after an emergency.
For telecommunicators serving in PSAPs, however, the changes will be more dramatic. New
specializations in addition to new data sources, analytics software, and dispatch workflows stand
to reshape emergency dispatch work. To considering these implications requires recognizing social
media as one or many potential information sources that will become available to PSAPs using
NG-911 infrastructure.
First, among first responders relying on information PSAPs dispatch, the impact of
incorporating social media monitoring in emergency response may be initially minimal, perhaps
even unnoticeable. Notably qualifying prior studies that find officials’ perceive social media as an
204
untrustworthy source of information (Hiltz & Plotnick, 2016; Plotnick et al., 2015; Reuter et al.,
2016), incorporating social media monitoring in emergency dispatch will not be immediately
apparent to first responders: they will continue to receive radio dispatches that, generally, never
attribute the source of the information. First responders, of course, now assume the information
they receive is relayed from 911 callers. However, his stands to change if analysts integrate
supplemental information from social media along with 911 caller information. How dispatchers
relay information reported on social media will likely depend on the nature of the emergency and
protocols developed as PSAPs begin incorporating indirect reports from social media in operational
decision-making. In lieu of real-world deployments, these dispatch protocols remain, for now,
unknown.
While first responders may not initially notice changes in how they receive information,
the extent to which social media provides useful, indirect reports of emergencies stands to affect
their interactions with the public. Whereas 911 typically brings first responders into contact with
citizens directly requesting help, indirect reports gleaned from social media stand to mediate
interactions between officials attempting to proactively manage risks and citizens unaware,
unconcerned, or contesting the very existence or nature of such risks. To consider these possibilities
requires addressing social media as one of many new information sources that stand to provide
indicators of emergencies in progress as well as risks associated with potential emergencies, e.g.
the assembly of large crowds, weather and traffic conditions, predictive analytics using historical
data. These changes will broadly impact “smart city” initiatives for civic services whose provision
will rest on remote data collection via physical and social sensors, (predictive) analytics using real-
time and historical data, and data-driven, operational decision-making (Caragliu, Del Bo, &
Nijkamp, 2011; Gabrys, 2014; Liu et al., 2015). Collectively, these initiatives employ or
hypothesize processes for the collection, analysis, and exploitation of data created by citizens that
205
will transform how civic services, including emergency services, are delivered and, of course, how
citizens will access and participate in the provision of such services.
For emergency dispatchers and administrators in the 5,783 PSAPs in the United States
(NENA, 2017), however, changes accompanying the prospective incorporation of social media will
be immediately apparent. These changes must be considered in light of the broader shifts that will
transform the PSAP from a reactive call center to a data analytics hub coordinating reactive and
proactive responses to emergencies and risks. These shifts will likely see i) new specializations
among PSAP staff and new procedures, policies, and training resources for officials assuming these
roles, ii) new workflows among dispatch teams, and iii) new sociotechnical infrastructures that can
enable and coordinate these workflows during various emergencies and potential disasters.
Incorporating new data sources such as social media in PSAPs will likely require new
specializations to complement the traditional call taker and dispatcher. Either by hiring new
personnel or re-training existing staff, those assuming the duties of “communications analyst” or
“social media analyst” will require new skills or adaptations of old ones. As the simulations of
Phase III reveal, telecommunicators’ experiences as call takers and dispatchers can inform and limit
the use of social media analysis tools. Expertise using CAD and the procedures that govern dispatch
work, as well as knowledge of the situational awareness information needs of first responders,
enable telecommunicators-turned-analysts to seamlessly participate in emergency dispatch
workflows. However, as the simulations of Phase III reveal, heuristics useful when questioning 911
callers, such as referring to nearby building names when attempting to ascertain a caller’s location,
do not always translate to effective search queries when searching social media data during an
emergency.
New specializations will require new procedures and training for individuals as well as
dispatch teams needing to cooperate with new staff, technologies, and information sources, as well
206
as the network of social media users who stand to participate, albeit indirectly, in emergency
dispatch operations. One implication of the role plays and simulations conducted with
telecommunicators is the need for new procedures, policies, and training materials in addition to
those provided to telecommunicators. While further simulations and operational experience will be
required to fully elaborate these requirements, prospective analysts will likely require substantial
expertise and training. Overall, significant questions still surround the incorporation of social media
analysts and analytics in emergency dispatch work (Table 6-1).
Table 6-1. Outstanding questions for incorporating social media in emergency dispatch.
What are the primary use cases for social media monitoring?
• What kinds of emergencies?
• What kinds of information during emergencies?
• Do PSAPs need analysts continually on call or during certain occasions only?
What social media analytics do PSAPs require for specific emergency situations?
• What kinds of automated filtering and alerts?
• What kinds of information seeking procedures?
How should analysts enter information discovered on social media in CAD systems?
• What source attributes should be provided to first responders?
How will collected social media data be stored?
• When? Where? For how long?
• Will the same policies apply for social media as for recording and storing 911 calls?
When do direct or indirect reports on social media suffice for dispatching emergency units?
• What types of emergencies?
• What types of messages (direct or indirect)?
• What information (e.g. corroborating posts) is required for verification?
Should analysts communicate directly with social media users during an emergency?
• If so, in what circumstances?
These questions are significant and require further research during deployments of social
media monitoring in simulated and actual PSAP operations. Collectively, the studies constituting
this dissertation contribute to the feasibility of such deployments by providing opportunities to
collect social media data within a geographic jurisdiction and articulating general procedural and
technical requirements for the inclusion of social media analysts and analytics in emergency
207
dispatch workflows. Further simulations can provide opportunities to refine these requirements for
specific emergency situations, formalize protocols for verifying information gathered from social
media users, and entering that information into CAD. These protocols will complement the basic
procedures for call taking and dispatch that provide the basis for telecommunicator training and
coordination among call takers, dispatchers, and first responders (APCO International, 2017).
Such research opportunities will arise as PSAPs begin trial deployments of social media
monitoring. Telecommunicators assuming the role of social media analysts (or, more broadly,
simply analysts) stand to play a crucial role as end users of social media analytics and participants
in research and design efforts connecting the hypothetical and actual. In this regard, research
approaches such as diary studies can facilitate participation among telecommunicators-turned
analysts in research addressing use cases for social media monitoring (e.g. which emergencies?
what information?) and provide usability feedback on social media analytics introduced into
emergency dispatch work.
In addition to telecommunicators, PSAP administrators will also play a crucial role as the
officials primarily responsible for negotiating the policy and staffing requirements for social media
analysts in emergency dispatch and response. Administrators will have to develop policies related
to the storage of social media data, nature of communications between the PSAP and social media
users, and the staffing of analyst positions. Researchers can contribute to policy development by
summarizing and sharing nascent policies developed among the many, autonomous community-
level jurisdictions across the United States (and internationally) and facilitating cooperation among
administrators and various stakeholders (professional associations, technology companies, social
media platforms, and community organizations) to develop policies that enhance emergency
response and are transparent and accountable to the public.
208
Implications for Theorizing “Actionability”
The field of Crisis Informatics consists of research examining uses of Information and
Communications Technologies (ICTs) during emergencies and disasters (Hagar, 2010; Palen et al.,
2009; Palen & Anderson, 2016). Since the field’s inception, social media has remained the cynosure
of Crisis Informatics research (Reuter, Hughes, & Kaufhold, 2018; Reuter & Kaufhold, 2018),
oriented by some variant of the question: what does social media contribute to situational
awareness? (Vieweg et al., 2010). While essential to the motivation and growth of the field, Crisis
Informatics has outgrown exclusive and, lately, mostly cursory employments of the orientation.
The field requires new theoretical guidance that brings together the findings of the past decade
while providing fertile grounding to examine emerging questions. Sensemaking provides a
complementary approach and, critically, provides grounding for not only studies of social media
per se, but the processes in which social media represents one of an array of information sources
citizens and emergency response officials rely on during crises. Sensemaking focuses attention on
the processes in which information gathered from multiple sources becomes useful, trustworthy,
and “actionable.”
Situational awareness, as defined by Endsley (1995), is a “state of knowledge” concerning
the perception of elements in the environment, comprehension of relations among elements, and
the projected status of these elements in the future. Appearing in Crisis Informatics studies of social
media, “situational awareness” provides motivation for content analyses of information people post
during emergencies and disasters (Olteanu et al., 2015; Starbird et al., 2010; Vieweg et al., 2010).
By examining the volume and diversity of information content posted by social media users
(Olteanu et al., 2015), researchers attempt to discover what social media may “may contribute to
situational awareness” (Vieweg et al., 2010). The necessary ambivalence of these studies’ findings
reflects the phenomena of investigation: information content vice states of current and projective
209
knowledge among citizens and emergency responders accessing information on social media
platforms.
Recently, noting that “focusing on supporting situational awareness (SA) may have some
weaknesses,” Zade and colleagues suggest an alternative approach of “actionability” (Zade et al.,
2018, p. 195:2):
The original situational awareness framing pushed us to filter relevant information from
irrelevant information and helped guide the categorizations of social media posts according
to different types of impacts (e.g., damage, human injury, evacuation notices). Now, the
actionability framing pushes us to consider further filtering and prioritizing—to figure out
what type of information is needed by a specific responder at a specific time. (195:16)
The orientation to actionability refines the situational updates that should be collected and
processed during a crisis to meet the varying requirements of emergency responders. Importantly,
both orientations constitute an attempt to map relationships between information content available
on social media, use contexts, and the situational awareness requirements of emergency responders.
This premise becomes problematic, however, when information can be “actionable” or
“unactionable” in similar use cases as a result of contextual factors, such as when information
available from existing sources makes information available on social media redundant, or when
“nonactionable” information becomes suddenly “actionable,” and vice versa.
While, at best, the concept of actionability highlights the contextual and processual features
of information relevance and utility which characterize emergency response work, at worst, the
concept restricts the scope of the problem to emergency responders’ need for bespoke systems
capable of providing personalized information outputs. Consequently, while the concept of
actionability can help refine the design of automated classification and filtering techniques, for the
goal of “integrating these systems into the practices of actual emergency responders” [italics
added], however, actionability does not describe the processes in which information not only
initiates but sustains action. Consequently, further attention needs to be given to the processes
210
through which information becomes actionable. To fully develop the concept of actionability
requires addressing the conditions that enable cooperation among officials (and citizens) engaged
in efforts to understand dynamic situations through the use of multiple ICTs and data sources.
Happily, in studies of Computer-Supported Cooperative Work, as well as Crisis Informatics, these
processes already have a well-established theoretical foundation: sensemaking.
Whereas actionability focuses on information for action, sensemaking focuses on
information in action. As previously defined in Phase III, sensemaking involves the deliberate
attempt to understand events by making sense of information while engaged in activities of
emergency response. The studies constituting this dissertation elaborate an approach to
actionability-as-sensemaking by examining, first, the types of relevant information that initiate
action among emergency dispatchers and, second, the processes of information in action wherein
dispatchers analyze, aggregate, and synthesize in which this information is made actionable for first
responders in the course of cooperative emergency response work.
First, regarding information initiating action, emergency dispatchers look for direct and
indirect reports of emergency that describe the 6Ws: Where, What, Weapons, Who, When, and
Why (Kropczynski et al., 2018). In this sense, actionable information describes the attributes and
sub-attributes of the 6Ws, the domain ontology emergency responders use to characterize an
emergency. The initial design workshop conducted at the Charleston County Consolidated 9-1-1
Center in August 2018 revealed the 6Ws as a simple, yet efficient, framework for understanding
information relevance in emergency dispatch and response work.
From the standpoint of sensemaking, Phase II introduces and evaluates data collection
methods that condition opportunities for automated filtering and user search techniques to identify
relevant information from the expanded volume and diversity of social media data collected using
location, keyword, and network filtering methods. The ongoing development of automated filtering
211
methods can leverage such data in systems designed to provide the kinds of relevant information
outputs that initiate action among emergency dispatchers. In this regard, spurred by the challenges
of information access that characterize social media monitoring in geographically-constrained
areas, the methods of Social Triangulation and network filtering contribute to infrastructure
necessary for incorporating social media in community emergency response.
Second, regarding information in action, Phase III employed role plays among dispatchers
to examine existing emergency dispatch workflows and simulations to examine future workflows
involving call takers, dispatchers, and social media analysts. Here relevant information gathered
from social media, that providing 6W information during an emergency, initiates workflows of
information integration that frame, elaborate, question, and reframe emergency situations
encountered during each simulation. Beyond initiating the actions constituting these workflows,
information gathered from social media sustains action by resourcing cooperative work among the
call taker, dispatcher, and analyst as they make sense of the situation using information from both
social media users and 911 callers. Beyond simply processing information gathered from social
media for dispatch to first responders, these processes, modeled as workflows consisting of
sequences of sensemaking activities (i.e. framing, elaborating, questioning and reframing),
articulate the conditions for relevance and actionability according to which the information outputs
of the analytics dashboard were assessed. When information gathered from 911 callers articulates
the scope of relevance and redundancy for subsequent information gathered from social media, the
simulations demonstrate how the “actionability” of information remains contingent on cooperative
processes integrating information from multiple data sources. In the workflows of call takers,
dispatchers, and analysts, relevant information discovered on social media not only initiates action
but, together with existing, alternative information sources, sustains actions that re-articulate
conditions for information seeking and retrieval in emergency dispatch and response.
212
As a result, approaching actionability from a sensemaking perspective encourages Crisis
Informatics research to move beyond studies focused exclusively on social media use and, instead,
examine the processes by which emergency responders gather, analyze, and integrate information
from heterogeneous data sources created by various physical and social sensors. These studies can
address two complimentary challenges. First, at what can be described as the information layer- the
array of information “outputs” available to officials including 911 calls, social media, first
responder radio traffic, camera imagery, etc., studies must address the cooperative and coordinative
work required for analyzing, identifying redundancy, synthesizing and aggregating information
supporting the situational needs of first responders. Phase III addressed this challenge by drawing
on the data-frame sensemaking theory of Klein and colleagues to identify elemental workflows of
information integration (Gary Klein et al., 2006, 2007). As discussed in Phase III, these studies can
attend to the coordination mechanisms enabling these workflows. In the context of the PSAP, these
include the domain ontology of the 6Ws, the common information space/operational picture of
CAD, and prospective information seeking procedures that will allow analysts to proactively
address information gaps articulated by call takers, and vice versa.
Second, the shift to sensemaking and information integration also encourages the design
of systems supporting sensemaking through data fusion. In contrast to information integration,
which occurs at the information layer and involves interactions between people and interfaces, data
fusion occurs at the data layer and involves systems processing data collected by physical and social
sensors (Hall & Llinas, 1997; Hall, Llinas, & Llinas, 2001). Here approaches for automated filtering
and summarization might synthesize information across multiple situational reports on social media
(Imran et al., 2015; Rudra et al., 2016), especially when information associated with the 6Ws is
likely to be distributed across multiple social media posts during an emergency. These summaries
213
can inform alert mechanisms and provide information for action: notifying the analyst an event has
occurred and supporting subsequent information seeking (Hall, Cai, & Graham, 2016).
Implications for Crisis Informatics Design Research
Finally, the methods guiding the phased studies of this dissertation demonstrate the value
of an iterative design approach that moves from analyses of existing activities to future activities.
In Phase I, interviews and scenario-based design methods helped articulate uses of social media in
the existing activity context of community emergency response and, at the same time, consider
future incorporations of social media distribution and monitoring. In Phase III, role play methods
revealed existing workflows and internal and external resources coordinating emergency dispatch,
as well as future workflows which arose when including an additional role— the social media
analyst— alongside call takers and dispatchers. Simulations further explored these future
workflows to highlight sociotechnical requirements related to analysts’ uses of the social media
analytics dashboard and the coordination of distributed sensemaking among the dispatch ensemble.
This approach addresses activity as the unit of analysis and unit of design (Kuijer, Jong, &
Eijk, 2008; Norman, 2005; Winograd et al., 1997). Activities (or practices) are socially-shared and
goal-directed ways of acting in the world (Kuutti & Bannon, 2014). Activities can be typical and
often mundane, but also constitute the majority of interactions of work and leisure. Activities are
socially-shared, often performed among people working cooperatively to accomplish an objective
and, as a result, involve ways of coordinating this work. This dissertation defines activities as sets
of purposeful (i.e. goal-directed), interdependent (i.e. cooperative, coordinative), and
transformative actions among people and artifacts.
214
Taking activity as the unit of analysis involves the analytic framing of existing activities to
describe objectives and needs, cooperative and coordinated work, and the products of these
interdependent actions. Phase I describes the existing activities of community emergency response
to highlight the objectives, cooperative actions, and coordinative requirements for incorporating
social media distribution and monitoring within this activity context. However, by employing
scenarios depicting prospective uses of social media, participating emergency responders also
consider future activities and the conditions that would enable them.
To take activity as the unit of design involves facilitating new configurations of people,
objects, and the meanings assigned to each, which, in turn, open opportunities to design artifacts
for future activities revealing themselves in and through these configurations. As Kuijer et al.
(2013) explain, “taking practices [i.e. activities] as a unit of design means to generate
reconfigurations of images [i.e. meanings], skills and stuff and their links, and requires the inclusion
of bodily performance, the creation of crises of routine and a variety of performances” (p. 21:19).
The emergent design of future activities through role plays and simulations in Phase III
reconfigured people, objects, and meanings through enactments that revealed opportunities to
design artifacts enabling these activities. Moreover, including social media analysts and analysis
tools in these role plays and simulations interrupted routine dispatch workflows and prompted
dispatchers to improvise actions to cooperate with analysts and coordinate the integration of
information gathered from social media users and 911 callers.
The use of “enactments” differs from, but complements, traditional usability studies in that
the object of design changes from the artifact in use to the activity of use. Enactments are often
discussed as an approach to speculative or future design. “User enactments,” explains Odom et al.
(2012) require “users to enact scenarios in which they get glimpses of several potential futures and
to use their own experiences to critically make sense of what they encountered” (p. 346). Elsewhere,
215
“speculative enactments,” according to Elsden et al. (2017), “constitute an effort to meaningfully
enact elements of possible futures with participants” (p. 5387). More generally, enactments appear
in experiential, scenario-based usability studies with low and hi-fidelity prototypes that employ
such methods as role play (Medler & Magerko, 2010; Simsarian, 2003; Svanaes & Seland, 2004),
improvisation (Kuijer et al., 2008; Sirkin & Ju, 2015), “prototyping social interaction” (Kurvinen,
Koskinen, & Battarbee, 2008), and “experience prototyping” (Buchenau & Suri, 2000). Here,
enactments refer to situations in which prospective end-users participate in the articulation of future
activities through uses of scenarios and artifacts introduced by design researchers.
Across the phased studies of the dissertation, gradually imposing constraints on enactments
allowed researchers and participating emergency response officials to focus on specific levels of
the activity context (i.e. activity, task, action, operation) and, within which, specific levels of system
design (i.e. hardware, interface, feature). On one hand, moving from low to high-fidelity
enactments represents the essential movement of design itself. In the framework of human-
centered design, this process begins with the analysis of human activities, moves to the ideation,
design, ad evaluation of artifacts, and, ultimately, the appropriation of these artifacts in now-
modified patterns of activity (Carroll & Rosson, 1992). Furthermore, the utility of low-fidelity
enactments in demonstrating the “what” (but not the “how”) of an existing or speculative activity
follows well-established approaches to design that focus attention on activities of use from the
earliest stage of the design process (Carroll, 2000; Rosson & Carroll, 2009).
In Phase III, role plays provided low-fidelity enactments in which telecommunicators were
free to improvise the “what” of future dispatch activities but not the “how” of systems hypothesized
in the role plays. The simulations, however, imposed significantly greater constraints in the form
of the environment, systems, and scenarios that focused telecommunicators on the “how” of using
software to search social media data and entering discovered information in CAD, etc. Moving
216
from low to high-fidelity enactments in this manner evoked sociotechnical requirements at multiple
levels of analysis, enabling the articulation of design requirements for both future activities and
systems enabling and constraining these activities.
By employing low and high-fidelity enactments, this dissertation demonstrated the utility
of selectively introducing constraints in ways that, first, focus on the design of activities and,
second, design for activities. First, this dissertation attempted the design of activities by outlining
emergency dispatch workflows, and sensemaking workflows in general. This was possible through
the recursive employment of low and high-fidelity enactments introducing loose and tight
constraints on action, respectively. Low-fidelity enactments in the form of role plays reconfigured
existing workflows to reveal new activities enabled by the inclusion of social media analysts and
analytics. Implications of the role plays for the design of and for activities, informed facilitation of
high-fidelity simulations that, in turn, revealed new actions (e.g. how analysts search social media)
and action-level requirements for artifacts, including assisted search features and proactive
information seeking protocols.
Second, this dissertation sought to design for future activities by articulating requirements
for the design of artifacts, including software such as social media analytics dashboards and analyst
workstations, as well as procedures instantiated, for example, in training resources for prospective
social media analysts. Phase III outlines requirements for such artifacts when discussing
sociotechnical requirements for incorporating social media analysts and analytics in emergency
dispatch workflows. These emerged in and through the low and high-fidelity enactments facilitated
by researchers and engaged in by telecommunicators. Low-fidelity, role play enactments revealed
activity-level requirements, including shared interpretive framework and domain ontology (i.e.
6Ws) and common information space (i.e. CAD). Simulations included additional constraints
additional constraints on action informed by the role plays and revealed action and operation-level
217
requirements for social media analysts (i.e. information seeking protocols, assisted-search features,
domain-specific visualizations).
Moving from low to high-fidelity enactments in this manner allowed design of and for
activities at progressively lower levels of analysis and design. Moving from low and hi-fidelity
enactments, from an interaction design perspective, coalesces established approaches facilitating
user participation to design for future activities (Elsden et al., 2017; Odom et al., 2012, 2014), and
suggests important applications for research and design in the field of Crisis Informatics.
Enactments can bridge divides that separate researchers designing social media analytics and
practitioners who need to effectively incorporate useful and useable systems in ways that transform
existing work practices to desired ones. That is, facilitating practitioner participation in low and
high-fidelity enactments provides people with resources in the form of experience, knowledge,
skills, and artifacts that can transform existing activities at the same time as informing the design
of artifacts that will enable these transformed, future activities. Enactments, then, function as design
methods to elicit requirements for future activities and artifacts, as well as design resources for
participants appropriating new artifacts and reconfiguring existing activities.
As such, studies deliberately moving from low to high-fidelity enactments can help
practitioners ideate future activities (using low-fidelity enactments) and iteratively define the types
of procedures and systems these activities will require and how they can be effectively employed
during a crisis (using high-fidelity enactments). In Phase III, dispatchers and PSAP administrator’s
participation in role plays and simulations became a resource for ongoing planning surrounding
staffing and coordinating the incorporation of social media analysts in future emergency dispatch
operations.
This becomes especially important in the field of Crisis Informatics where design work is
typically restricted from emergency and disaster use contexts where design activities with
218
emergency responders-as-end users is impossible and inappropriate. Moreover, as crises remain
highly unpredictable and locally-contingent (“all disasters are local”), the design of and for future
activities takes on greater urgency. For these reasons, design studies can engage emergency
responders in low and high-fidelity enactments that enable the design of future activities and future
artifacts at different levels of analysis, and, through participation in enactments themselves,
facilitate the participatory design and appropriation of these activities and artifacts in future
contexts of use.
Directions for Future Research
Although this dissertation focuses on the incorporation of social media in community
emergency response, its methodology provides an approach to support the transition of PSAPs from
reactive call centers to proactive data analytics centers collecting, processing, and synthesizing data
from multiple sources. As NG911 infrastructure will enable PSAPs to receive text, images, and
streaming video, new procedures and analysis tools will be necessary to process data from these
sources and make information actionable for first responders. Many of the same problems facing
PSAPs attempting to incorporate social media arise when incorporating multimedia data: PSAPs
require new activities and workflows enabled by new tools and procedures.
Methodologically, enactments can allow researchers and PSAP staff to co-design future
activities for processing multimedia data and the protocols and systems that will enable these
activities. Furthermore, enactments can, themselves, assist PSAPs transition to NG911
infrastructure by providing training and troubleshooting exercises for PSAP administrators, call
takers, dispatchers, new communications specialists, as well as first responders who will be the
ultimate end-users of multimedia information. As previously described, moving from low-fidelity
219
enactments, focused on the design of activities, to high-fidelity enactments, focused on protocols
and artifacts enabling those activities, provides an integrative and holistic participatory design
approach that can help transition legacy systems and associated standard operating procedures to
future work practices. In the case of community emergency response, this transition is motivated
by the disruptions caused by multimedia information sharing on mobile devices and sociotechnical
affordances arising with the widespread use of social media platforms and IP-based
communications infrastructure.
Moving to NG911 dispatch workflows in which call takers and dispatchers will receive
and make sense of text, images, and video will disrupt the normal work of emergency dispatch.
Challenges include technical solutions to send image or video to first responders’ mobile data
terminals as well as the aggregation and visualization of multimedia information for efficient
sensemaking among dispatch teams. Incorporating social media, including images and videos
posted on social media platforms, raises similar challenges. To understand these future workflows,
low-fidelity enactments, similar to those in with dispatchers participated in Phase III, can allow
emergency response officials to walk through scenarios in which PSAPs expect to encounter
multimedia information and scenarios in which first responders would like to receive multimedia
information. Low-fidelity enactments such as role plays with limited or no props allow dispatch
staff and first responders to improvise what they would do in ways that expand the design space by
elaborating general affordances (e.g. 911 callers can send images to call takers) in concrete
situations, identifying likely contingencies, and developing ad hoc protocols and best practices.
Activities revealed through low-fidelity enactments provide the context for exploring and
specifying design requirements for NG911 protocols and systems needed to process multimedia
data. Similar to the simulations conducted in Phase III, high-fidelity enactments with emergency
dispatchers can reveal how they need to collect, process, and relay multimedia information to first
220
responders. By closely simulating the conditions of emergency dispatch and response, high-fidelity
enactments provide opportunities for usability analyses that can inform the design of protocols and
artifacts that will be required with the progressive implementation of NG911 infrastructure. As
demonstrated in the studies comprising this dissertation, progressively employing low and then
high-fidelity enactments provides opportunities for the design of and for activities of next-
generation emergency dispatch and response.
Lastly, design enactments provide resources for the transition from legacy systems and
associated work practices to new systems and processes such as those envisioned in NG911.
Enactments in the form of training exercises, for instance, provide PSAP administrators,
dispatchers (including communications specialists), and first responders with shared experiences,
skills, and nascent procedures for appropriating new systems allowing multimedia communications
with citizens. However, as officials remain occupied with existing duties and often lack the
requisite resources to conduct such exercises, the ability for researchers and practitioners to
effectively collaborate likely figures as a precondition for effective and rapid innovation in public
safety work. Again, the collaborative work conducted in this dissertation provides a model for
understanding the challenges of designing and deploying NG911 systems in community emergency
response. To address these challenges, the applied, methodological, and theoretical contributions
of this dissertation offer a promising foundation.
221
References
911.Gov. (2016). Next Generation 911 for leaders in Law Enforcement. Retrieved from
https://www.911.gov/project_ng911publicsafety/law/digital-edition.html
Agar, M. (1986). Speaking of Ethnography. Newbury Park, CA: Sage.
Ajao, O., Hong, J., & Liu, W. (2015). A survey of location inference techniques on Twitter. Journal
of Information Science, 41(6). https://doi.org/10.1177/0165551515602847
Aldrich, D. P., & Meyer, M. A. (2015). Social Capital and Community Resilience. American
Behavioral Scientist, 59(2), 254–269. https://doi.org/10.1177/0002764214550299
Amir, S., & Kant, V. (2017). Sociotechnical Resilience: A Preliminary Concept. Risk Analysis.
https://doi.org/10.1111/risa.12816
Annarelli, S., & Hartley, J. (2017, May 2). Power out, schools closed after Tuesday storm. Centre
Daily Times. Retrieved from
http://www.centredaily.com/news/local/community/article148010634.html
Anson, S., Watson, H., Wadhwa, K., & Metz, K. (2017). Analysing social media data for disaster
preparedness: Understanding the opportunities and barriers faced by humanitarian actors.
International Journal of Disaster Risk Reduction, 21, 131–139.
https://doi.org/10.1016/J.IJDRR.2016.11.014
APCO International. (2017). Public Safety Telecommunicator (PST), 7th Edition. Retrieved
February 6, 2019, from https://www.apcointl.org/training-and-
certification/disciplines/public-safety-telecommunicator-pst/public-safety-
telecommunicator/
Avvenuti, M., Cresci, S., Marchetti, A., Meletti, C., & Tesconi, M. (2014). EARS (Earthquake
Alert and Report System): A Real Time Decision Support System for Earthquake Crisis
Management. Proceedings of the 20th ACM SIGKDD International Conference on
Knowledge Discovery and Data Mining, 1749–1758.
https://doi.org/10.1145/2623330.2623358
Baber, C., & McMaster, R. (2016). Grasping the Moment: Sensemaking in Response to Routine
Incidents and Major Emergencies. Boca Raton, FL: CRC Press.
Backstrom, L., Sun, E., & Marlow, C. (2010). Find me if you can. In Proceedings of the 19th
international conference on World wide web - WWW ’10 (p. 61). New York, New York, USA:
ACM Press. https://doi.org/10.1145/1772690.1772698
Bannon, L., & Bødker, S. (1997). Constructing Common Information Spaces. In Proceedings of
the Fifth European Conference on Computer Supported Cooperative Work (pp. 81–96).
Dordrecht: Springer Netherlands. https://doi.org/10.1007/978-94-015-7372-6_6
222
Bauman, K., Tuzhilin, A., & Zaczynski, R. (2017). Using Social Sensors for Detecting Emergency
Events. ACM Transactions on Management Information Systems, 8(2–3), 1–20.
https://doi.org/10.1145/3052931
Bean, H., Sutton, J., Liu, B. F., Madden, S., Wood, M. M., & Mileti, D. S. (2015). The Study of
Mobile Public Warning Messages: A Research Review and Agenda. Review of
Communication, 15(1), 60–80. https://doi.org/10.1080/15358593.2015.1014402
Beneito-Montagut, R., Anson, S., Shaw, D., & Brewster, C. (2013). Governmental Social Media
use for Emergency Communication. In Proceedings of the 10th International ISCRAM
Conference (pp. 1–5). Baden-Baden, Germany. Retrieved from
http://www.cbrewster.com/papers/Beneito-Montagut_ISCRAM13.pdf
Boehner, K., Vertesi, J., Sengers, P., & Dourish, P. (2007). How HCI interprets the probes. In
Proceedings of the SIGCHI conference on Human factors in computing systems - CHI ’07
(p. 1077). New York, New York, USA: ACM Press.
https://doi.org/10.1145/1240624.1240789
Boersma, K., Diks, D., Ferguson, J., & Wolbers, J. (2016). From Reactive to Proactive Use of
Social Media in Emergency Response: A Critical Discussion of the Twitcident Project. In
Emergency and Disaster Management: Concepts, Methodologies, Tools, and Applications
(pp. 236–252). IGI Global. https://doi.org/10.4018/978-1-4666-9867-3.ch014
Bruns, A., Burgess, J., Crawford, K., & Shaw, F. (2012). #qldfloods and @QPSMedia: Crisis
Communication on Twitter in the 2011 South East Queensland Floods. Brisbane QLD
Australia.
Bruns, A., & Liang, Y. E. (2012). Tools and methods for capturing Twitter data during natural
disasters. First Monday, 17(4). https://doi.org/10.5210/fm.v17i4.3937
Bryant, A., & Charmaz, K. (2007). Grounded Theory in historical perspective: An epistemological
account. In A. Bryant & K. Charmaz (Eds.), The SAGE handbook of grounded theory (pp.
31–58). London: Sage.
Brynielsson, J., Granåsen, M., Lindquist, S., Narganes Quijano, M., Nilsson, S., & Trnka, J. (2018).
Informing crisis alerts using social media: Best practices and proof of concept. Journal of
Contingencies and Crisis Management, 26(1). https://doi.org/10.1111/1468-5973.12195
Buchenau, M., & Suri, J. F. (2000). Experience prototyping. In Proceedings of the conference on
Designing interactive systems processes, practices, methods, and techniques - DIS ’00 (pp.
424–433). New York, New York, USA: ACM Press. https://doi.org/10.1145/347642.347802
Cagle, M. (2016). Facebook, Instagram, and Twitter Provided Data Access for a Surveillance
Product Marketed to Target Activists of Color. Retrieved September 18, 2017, from
https://www.aclunc.org/blog/facebook-instagram-and-twitter-provided-data-access-
surveillance-product-marketed-target
CajunNavy. (2017). If you or someone you know needs help [Facebook update]. Retrieved from
223
https://www.facebook.com/LaCajunNavy/
Cameron, M. A., Power, R., Robinson, B., & Yin, J. (2012). Emergency situation awareness from
twitter for crisis management. In Proceedings of the 21st international conference companion
on World Wide Web - WWW ’12 Companion (p. 695). New York, New York, USA: ACM
Press. https://doi.org/10.1145/2187980.2188183
Caragliu, A., Del Bo, C., & Nijkamp, P. (2011). Smart Cities in Europe. Journal of Urban
Technology, 18(2), 65–82. https://doi.org/10.1080/10630732.2011.601117
Carley, K. M., Malik, M., Landwehr, P. M., Pfeffer, J., & Kowalchuck, M. (2016). Crowd sourcing
disaster management: The complex nature of Twitter usage in Padang Indonesia. Safety
Science, 90, 48–61. https://doi.org/10.1016/J.SSCI.2016.04.002
Carroll, J. M. (2000). Making Use: Scenario-Based Design of Human-Computer Interactions.
Cambridge, MA: MIT Press.
Carroll, J. M. (2015). Theorizing the unprecedented. In D. H. Sonnenwald (Ed.), Theory
development in the information sciences (pp. 185–203). University of Texas Press.
Carroll, J. M., & Kellogg, W. (1989). Artifact as theory-nexus: hermeneutics meets theory-based
design. In Proceedings of the SIGCHI conference on Human factors in computing systems
Wings for the mind - CHI ’89 (Vol. 20, pp. 7–14). New York, New York, USA: ACM Press.
https://doi.org/10.1145/67449.67452
Carroll, J. M., & Rosson, M. B. (1992). Getting around the task-artifact cycle: how to make claims
and design by scenario. ACM Transactions on Information Systems, 10(2), 181–212.
https://doi.org/10.1145/146802.146834
Carroll, J. M., & Rosson, M. B. (2007). Participatory design in community informatics. Design
Studies, 28(3), 243–261. https://doi.org/10.1016/J.DESTUD.2007.02.007
Carroll, J. M., Shih, P. C., & Kropczynski, J. (2004). Community informatics as innovation in
sociotechnical infrastructures. The Journal of Community Informatics, 11(2). Retrieved from
http://ci-journal.org/index.php/ciej/article/view/1153
Chandra, A., Acosta, J., Howard, S., Uscher-Pines, L., Williams, M., Yeung, D., … Meredith, L.
S. (2011). Building Community Resilience to Disasters: A Way Forward to Enhance National
Health Security. Rand Health Quarterly, 1(1), 6. Retrieved from
http://www.ncbi.nlm.nih.gov/pubmed/28083162
Chatfield, A., & Brajawidaga, U. (2012). Twitter tsunami early warning network : a social network
analysis of Twitter information flows, 2012, 1–10.
Chauhan, A., & Hughes, A. L. (2017). Providing Online Crisis Information. In Proceedings of the
2017 CHI Conference on Human Factors in Computing Systems - CHI ’17 (pp. 3151–3162).
New York, New York, USA: ACM Press. https://doi.org/10.1145/3025453.3025627
224
Cheng, Z., Caverlee, J., & Lee, K. (2010). You are where you tweet. In Proceedings of the 19th
ACM international conference on Information and knowledge management - CIKM ’10 (p.
759). New York, New York, USA: ACM Press. https://doi.org/10.1145/1871437.1871535
Cobb, C., McCarthy, T., Perkins, A., Bharadwaj, A., Comis, J., Do, B., & Starbird, K. (2014).
Designing for the deluge. In Proceedings of the 17th ACM conference on Computer supported
cooperative work & social computing - CSCW ’14 (pp. 888–899). New York, New York,
USA: ACM Press. https://doi.org/10.1145/2531602.2531712
Compton, R., Jurgens, D., & Allen, D. (2014). Geotagging one hundred million Twitter accounts
with total variation minimization. In 2014 IEEE International Conference on Big Data (Big
Data) (pp. 393–401). IEEE. https://doi.org/10.1109/BigData.2014.7004256
Crampton, J. W., Graham, M., Poorthuis, A., Shelton, T., Stephens, M., Wilson, M. W., & Zook,
M. (2013). Beyond the geotag: situating ‘big data’ and leveraging the potential of the geoweb.
Cartography and Geographic Information Science, 40(2), 130–139.
https://doi.org/10.1080/15230406.2013.777137
Creswell, J. (2009). Research design: Qualitative, Quantitative, and mixed methods approaches.
Thousand Oaks, CA: Sage.
Cuff, D., Hansen, M., & Kang, J. (2008). Urban sensing. Communications of the ACM, 51(3), 24–
33. https://doi.org/10.1145/1325555.1325562
Cutter, S. L., Ahearn, J. A., Amadei, B., Crawford, P., Eide, E. A., Galloway, G. E., … Zoback, M.
Lou. (2013). Disaster Resilience: A National Imperative. Environment: Science and Policy
for Sustainable Development, 55(2), 25–29. https://doi.org/10.1080/00139157.2013.768076
Dailey, D., & Starbird, K. (2017). Social Media Seamsters: Stitching Platforms & Audiences
into Local Crisis Infrastructure. In Computer Supported Cooperative Work (CSCW).
https://doi.org/10.1145/2998181.2998290
Davis Jr., C., Pappa, G. L., Oliveira, D., & Arcanjo, F. (2011). Inferring the Location of Twitter
Messages Based on User Relationships. Transactions in GIS, 15(6), 735–751.
https://doi.org/10.1111/j.1467-9671.2011.01297.x
de Albuquerque, J. P., Herfort, B., Brenning, A., & Zipf, A. (2015). A geographic approach for
combining social media and authoritative data towards identifying useful information for
disaster management. International Journal of Geographical Information Science, 29(4),
667–689. https://doi.org/10.1080/13658816.2014.996567
Denef, S., Bayerl, P. S., & Kaptein, N. A. (2013). Social media and the police. In Proceedings of
the SIGCHI Conference on Human Factors in Computing Systems - CHI ’13 (p. 3471). New
York, New York, USA: ACM Press. https://doi.org/10.1145/2470654.2466477
DHS. (2008). National Incident Management System. Retrieved from
https://www.fema.gov/pdf/emergency/nims/NIMS_core.pdf
225
Drabek, T. (1986). Human System Responses to Disaster: An Inventory of Sociological Findings.
New York: Springer-Verlag. https://doi.org/10.1007/978-1-4612-4960-3
Drabek, T. E., & McEntire, D. A. (2003). Emergent phenomena and the sociology of disaster:
lessons, trends and opportunities from the research literature. Disaster Prevention and
Management: An International Journal, 12(2), 97–112.
https://doi.org/10.1108/09653560310474214
Easterby-Smith, M., Golden-Biddle, K., & Locke, K. (2008). Working With Pluralism.
Organizational Research Methods, 11(3), 419–429.
https://doi.org/10.1177/1094428108315858
Elsden, C., Chatting, D., Durrant, A. C., Garbett, A., Nissen, B., Vines, J., & Kirk, D. S. (2017).
On Speculative Enactments. In Proceedings of the 2017 CHI Conference on Human Factors
in Computing Systems - CHI ’17 (pp. 5386–5399). New York, New York, USA: ACM Press.
https://doi.org/10.1145/3025453.3025503
Endsley, M. R. (1995). Toward a Theory of Situation Awareness in Dynamic Systems. Human
Factors: The Journal of the Human Factors and Ergonomics Society, 37(1), 32–64.
https://doi.org/10.1518/001872095779049543
FEMA. (2007). Basic Guidance for Public Information Officers (PIOs) (FEMA 517). Retrieved
from https://www.fema.gov/media-library-data/20130726-1623-20490-
0276/basic_guidance_for_pios_final_draft_12_06_07.pdf
FEMA. (2011). A Whole Community approach to emergency management: Principles, themes,
and pathways for action. Retrieved from https://www.fema.gov/media-library-
data/20130726-1813-25045-0649/whole_community_dec2011__2_.pdf
FEMA. (2013). IS-702.A - NIMS Public Information (training course). Retrieved from
https://emilms.fema.gov/IS702A/PIOsummary.htm
Finch, J. (1987). The Vignette Technique in Survey Research. Sociology, 21(1), 105–114.
https://doi.org/10.1177/0038038587021001008
Furniss, D., & Blandford, A. (2006). Understanding emergency medical dispatch in terms of
distributed cognition: a case study. Ergonomics, 49(12–13), 1174–1203.
https://doi.org/10.1080/00140130600612663
Gabrys, J. (2014). Programming environments: environmentality and citizen sensing in the smart
city. Environment and Planning D: Society and Space, 32(1), 30–48.
https://doi.org/10.1068/d16812
Glaser, B. (1978). Theoretical sensitivity: Advances in the methodology of grounded theory. San
Francisco: Sociology Press.
Glaser, B., & Strauss, A. (1967). The discovery of grounded theory: Strategies for qualitative
research. Hawthorne, NY: Aldine de Gruyter.
226
Golafshani, N. (1990). Understanding reliability and validity in qualitative research. The
Qualitative Report, 8(4). Retrieved from https://nsuworks.nova.edu/tqr/vol8/iss4/6
Golden-Biddle, K., & Locke, K. (1993). Appealing work: An investigation of how ethnographic
texts convince. Organization Science, 4(4), 595–616. https://doi.org/10.1287/orsc.4.4.595
Gomez, E. A., & Passerini, K. (2004). Information and Communication Technologies (ICT)
Options for Local and Global Communities in Health-Related Crisis Management. The
Journal of Community Informatics, 6(2). Retrieved from http://ci-
journal.org/index.php/ciej/article/view/391
Gow, G., Waidyanatha, N., & Anderson, P. (2007). Community-Based Hazard Warnings in Rural
Sri Lanka: Performance of Alerting and Notification in a Last-Mile Message Relay. SSRN
Electronic Journal. https://doi.org/10.2139/ssrn.1572329
Grace, R., Halse, S., Aurite, W., Montarnal, A., & Tapia, A. (2019). Expanding awareness:
Comparing location, keyword, and network filtering methods to collect hyperlocal social
media data. In HICSS 2019.
Grace, R., Kropczynski, J., Obeysekare, E., Halse, S., Coche, J., Montarnal, A., … Beagles, M.
(2019). Role playing Next Generation 9-1-1: Sensemaking with social media in Public-Safety
Answering Points. In HICSS 2019.
Grace, R., Kropczynski, J., Pezanowski, S., Halse, S., Umar, P., & Tapia, A. (2019). Enhancing
emergency communication with social media: Identifying hyperlocal social media users and
information sources. International Journal of Information Systems for Crisis Response and
Management.
Grace, R., Kropczynski, J., Pezanowski, S., Halse, S., Umar, P., & Tapia, A. (2017). Social
Triangulation: A new method to identify local citizens using social media and their local
information curation behaviors. Proceedings of the 14th International Conference on
Information Systems for Crisis Response and Management, (May 2017), 902–915.
Grace, R., Kropczynski, J., & Tapia, A. (2018). Community coordination: Aligning social media
use in community emergency management. In K. Boersma & B. Tomaszewski (Eds.),
Proceedings of the 15th ISCRAM Conference. Rochester, NY.
Graham, M. W., Avery, E. J., & Park, S. (2015). The role of social media in local government crisis
communications. Public Relations Review, 41(3), 386–394.
https://doi.org/10.1016/J.PUBREV.2015.02.001
Greitzer, F. L., Schur, A., Paget, M., & Guttromson, R. T. (2008). A sensemaking perspective on
situation awareness in power grid operations. In 2008 IEEE Power and Energy Society
Meeting (pp. 1–6). IEEE. https://doi.org/10.1109/PES.2008.4596285
Guba, E., & Lincoln, Y. (1989). Fourth Generation Evaluation. Newbury Park, CA: Sage.
227
Gurstein, M. (2005). Tsunami warning systems and the last mile. The Journal of Community
Informatics, 1(2), 14–17. Retrieved from http://ci-journal.org/index.php/ciej/article/view/229
Gurstein, M. (2007). What is community informatics (and why does it matter)? Milan: Polimetrica.
Retrieved from https://arxiv.org/ftp/arxiv/papers/0712/0712.3220.pdf
Hagar, C. (2010). Introduction to the Special Section. Bulletin of the American Society for
Information Science and Technology, 36(5), 10–12.
https://doi.org/10.1002/bult.2010.1720360504
Hall, D., Cai, G., & Graham, J. (2016). Empowering the next-generation analyst. In G. Rogova &
P. Scott (Eds.), Fusion methodologies in crisis management: Higher level fusion and decision
making (pp. 231–244). Springer, Cham.
Hall, D. L., & Llinas, J. (1997). An introduction to multisensor data fusion. Proceedings of the
IEEE, 85(1), 6–23. https://doi.org/10.1109/5.554205
Hall, D., Llinas, J., & Llinas, J. (2001). Multisensor Data Fusion. (D. Hall & J. Llinas, Eds.). Boca
Raton: CRC Press. https://doi.org/10.1201/9781420038545
Halse, S. E., Tapia, A., Squicciarini, A., & Caragea, C. (2018). An emotional step toward automated
trust detection in crisis social media. Information, Communication & Society, 21(2), 288–305.
https://doi.org/10.1080/1369118X.2016.1272618
Hecht, B., Hong, L., Suh, B., & Chi, E. H. (2011). Tweets from Justin Bieber’s heart. In
Proceedings of the 2011 annual conference on Human factors in computing systems - CHI
’11 (p. 237). New York, New York, USA: ACM Press.
https://doi.org/10.1145/1978942.1978976
Hecht, B., & Stephens, M. (2014). A Tale of Cities: Urban Biases in Volunteered Geographic
Information. In Eighth International AAAI Conference on Weblogs and Social Media
(ICWSM) (pp. 197–205). Retrieved from
https://www.aaai.org/ocs/index.php/ICWSM/ICWSM14/paper/viewFile/8114/8120
Heidegger, M. (2010). Being and Time. Albany: SUNY Press.
Hernantes, J., Labaka, L., Turoff, M., Hiltz, S. R., & Bañuls, V. A. (2017). Moving forward to
disaster resilience: Perspectives on increasing resilience for future disasters. Technological
Forecasting and Social Change, 121, 1–6.
https://doi.org/10.1016/J.TECHFORE.2017.05.011
Hiltz, S. R., Kushman, J., & Plotnick, L. (2014). Use of Social Media by U.S. Public Sector
Emergency Managers: Barriers and Wish Lists. In Proceedings of the 11th International
ISCRAM Conference. Retrieved from
http://ww.w.iscram.org/legacy/ISCRAM2014/papers/p11.pdf
Hiltz, S. R., & Plotnick, L. (2013). Dealing with information overload when using social media for
emergency management: emerging solutions. In T. Comes, F. Fiedrich, S. Fortier, J.
228
Geldermann, & T. Mülle (Eds.), Proceedings of the 10th International ISCRAM Conference.
Baden-Baden, Germany.
Hiltz, S. R., & Plotnick, L. (2016). Barriers to Use of Social Media by Emergency Managers.
Journal of Homeland Security and Emergency Management, 13(2), 6–13.
https://doi.org/10.1515/jhsem-2015-0068
Holland, L. (2018). Charleston County public can now connect with 911 via web - WCBD.
Retrieved June 8, 2018, from http://www.counton2.com/news/local-news/charleston-county-
public-can-now-connect-with-911-via-web/1128896475
Hong, L., Fu, C., Torrens, P., & Frias-Martinez, V. (2017). Understanding Citizens’ and Local
Governments’ Digital Communications During Natural Disasters. In Proceedings of the 2017
ACM on Web Science Conference - WebSci ’17 (pp. 141–150). New York, New York, USA:
ACM Press. https://doi.org/10.1145/3091478.3091502
Houston, J. B., Hawthorne, J., Perreault, M. F., Park, E. H., Goldstein Hode, M., Halliwell, M. R.,
… Griffith, S. A. (2015). Social media and disasters: a functional framework for social media
use in disaster planning, response, and research. Disasters, 39(1), 1–22.
https://doi.org/10.1111/disa.12092
Hsieh, H.-F., & Shannon, S. E. (2005). Three Approaches to Qualitative Content Analysis.
Qualitative Health Research, 15(9), 1277–1288. https://doi.org/10.1177/1049732305276687
Huang, Y., Huo, S., Yao, Y., Chao, N., Wang, Y., Grygiel, J., & Sawyer, S. (2016). Municipal
Police Departments on Facebook. In Proceedings of the 17th International Digital
Government Research Conference on Digital Government Research - dg.o ’16 (pp. 366–374).
New York, New York, USA: ACM Press. https://doi.org/10.1145/2912160.2912189
Huang, Q., & Xiao, Y. (2015). Geographic Situational Awareness: Mining Tweets for Disaster
Preparedness, Emergency Response, Impact, and Recovery. ISPRS International Journal of
Geo-Information, 4(3), 1549–1568. https://doi.org/10.3390/ijgi4031549
Huang, Y., Wu, F., & Cheng, Y. (2016). Crisis communication in context: Cultural and political
influences underpinning Chinese public relations practice. Public Relations Review, 42(1),
201–213. https://doi.org/10.1016/J.PUBREV.2015.11.015
Hughes, A. (2014). Participatory Design for the Social Media Needs of Emergency Public
Information Officers. In Proceedings of the 11th International ISCRAM Conference (pp. 727–
736). University Park, PA. Retrieved from
http://live.iscram.org/legacy/ISCRAM2014/papers/p90.pdf
Hughes, A., & Palen, L. (2009). Twitter adoption and use in mass convergence and emergency
events. International Journal of Emergency Management, 6(3/4), 248.
https://doi.org/10.1504/IJEM.2009.031564
Hughes, A., & Palen, L. (2012). The Evolving Role of the Public Information Officer: An
Examination of Social Media in Emergency Management. Journal of Homeland Security and
Emergency Management, 9(1). https://doi.org/10.1515/1547-7355.1976
229
Hughes, A., & Shah, R. (2016). Designing an Application for Social Media Needs in Emergency
Public Information Work. In Proceedings of the 19th International Conference on Supporting
Group Work - GROUP ’16 (pp. 399–408). New York, New York, USA: ACM Press.
https://doi.org/10.1145/2957276.2957307
Hughes, A., & Tapia, A. H. (2015). Social Media in Crisis: When Professional Responders Meet
Digital Volunteers. Journal of Homeland Security and Emergency Management, 12(3), 679–
706. https://doi.org/10.1515/jhsem-2014-0080
Hughes, A. L., St. Denis, L. A. A., Palen, L., Anderson, K. M., Hughes, A. L., St. Denis, L. A. A.,
… Anderson, K. M. (2014). Online public communications by police & fire services
during the 2012 Hurricane Sandy. In Proceedings of the 32nd annual ACM conference on
Human factors in computing systems - CHI ’14 (pp. 1505–1514). New York, New York,
USA: ACM Press. https://doi.org/10.1145/2556288.2557227
Imran, M., Castillo, C., Diaz, F., & Vieweg, S. (2015). Processing Social Media Messages in Mass
Emergency. ACM Computing Surveys, 47(4), 1–38. https://doi.org/10.1145/2771588
Imran, M., Elbassuoni, S., Castillo, C., Diaz, F., & Meier, P. (2013a). Extracting Information
Nuggets from Disaster-Related Messages in Social Media. In Proceedings of the 10th
International ISCRAM Conference (pp. 1–10). Baden-Baden, Germany. Retrieved from
http://www.crowdflower.com
Imran, M., Elbassuoni, S., Castillo, C., Diaz, F., & Meier, P. (2013b). Practical extraction of
disaster-relevant information from social media. In Proceedings of the 22nd International
Conference on World Wide Web - WWW ’13 Companion (pp. 1021–1024). New York, New
York, USA: ACM Press. https://doi.org/10.1145/2487788.2488109
Jaspers, M. W. M., Steen, T., Van Den Bos, C., & Geenen, M. (2004). The think aloud method: a
guide to user interface design. International Journal of Medical Informatics, 73, 781–795.
https://doi.org/10.1016/j.ijmedinf.2004.08.003
Johnson, I. L., Sengupta, S., Schöning, J., & Hecht, B. (2016). The Geography and Importance of
Localness in Geotagged Social Media. In Proceedings of the 2016 CHI Conference on Human
Factors in Computing Systems - CHI ’16 (pp. 515–526). New York, New York, USA: ACM
Press. https://doi.org/10.1145/2858036.2858122
Jurgens, D. (2013). That’s What Friends Are For: Inferring Location in Online Social Media
Platforms Based on Social Relationships. In Proceedings of the Seventh International AAAI
Conference on Weblogs and Social Media (pp. 273–282). Retrieved from
https://www.aaai.org/ocs/index.php/ICWSM/ICWSM13/paper/viewFile/6067/6366
Jurgens, D., Finethy, T., Mccorriston, J., Xu, Y. T., & Ruths, D. (2015). Geolocation Prediction in
Twitter Using Social Networks: A Critical Analysis and Review of Current Practice. In
Proceedings of the Ninth International AAAI Conference on Web and Social Media (pp. 188–
197). Retrieved from
https://www.aaai.org/ocs/index.php/ICWSM/ICWSM15/paper/viewFile/10584/10502
230
Kaewkitipong, L., Chen, C., & Ractham, P. (2012). Lessons learned from the use of social media
in combating a crisis: A case study of 2011 Thailand flooding disaster. ICIS 2012
Proceedings. Retrieved from
https://aisel.aisnet.org/icis2012/proceedings/ProjectManagement/8
Kaptelinin, V., & Nardi, B. (2012). Activity Theory in HCI: Fundamentals and Reflections.
Synthesis Lectures on Human-Centered Informatics, 5(1), 1–105.
https://doi.org/10.2200/S00413ED1V01Y201203HCI013
Karanasios, S., Cooper, V., Balcell, M. P., & Hayes, P. (2019). Inter-Organizational Collaboration,
Information Flows, and the Use of Social Media During Disasters: A Focus on Vulnerable
Communities. Retrieved from https://scholarspace.manoa.hawaii.edu/handle/10125/59736
Kaufhold, M.-A., Gizikis, A., Reuter, C., Habdank, M., & Grinko, M. (2018). Avoiding chaotic
use of social media before, during, and after emergencies: Design and evaluation of citizens’
guidelines. Journal of Contingencies and Crisis Management. https://doi.org/10.1111/1468-
5973.12249
Kaufhold, M.-A., & Reuter, C. (2016). The Self-Organization of Digital Volunteers across Social
Media: The Case of the 2013 European Floods in Germany. Journal of Homeland Security
and Emergency Management, 13(1), 137–166. https://doi.org/10.1515/jhsem-2015-0063
Kaufhold, M. A., & Reuter, C. (2017). The Impact of Social Media for Emergency Services: A
Case Study with the Fire Department Frankfurt. In Proceedings of the 14th ISCRAM
Conference . Albi, France.
Kavanaugh, A. L., Fox, E. A., Sheetz, S. D., Yang, S., Li, L. T., Shoemaker, D. J., … Xie, L.
(2012). Social media use by government: From the routine to the critical. Government
Information Quarterly, 29(4), 480–491. https://doi.org/10.1016/J.GIQ.2012.06.002
Kerka, S. (2003). Community asset mapping. Clearinghouse on Adult, Career, and Vocational
Education, 47, 1–2. Retrieved from
http://srdc.msstate.edu/publications/227/227_asset_mapping.pdf
Klein, G., Moon, B., & Hoffman, R. R. (2006). Making Sense of Sensemaking 1: Alternative
Perspectives. IEEE Intelligent Systems, 21(4), 70–73. https://doi.org/10.1109/MIS.2006.75
Klein, G., Moon, B., & Hoffman, R. R. (2006). Making Sense of Sensemaking 2: A Macrocognitive
Model. IEEE Intelligent Systems, 21(5). Retrieved from www.computer.org/intelligent
Klein, G., Phillips, J., Rall, E., & Peluso, D. (2007). A Data-Frame Theory of Sensemaking. In R.
Hoffman (Ed.), Expertise Out of Context: Proceedings of the Sixth International Conference
on Naturalistic Decision Making (pp. 113–158). New York: Lawrence Erlbaum Associates.
Klein, H. K., & Myers, M. D. (1999). A Set of Principles for Conducting and Evaluating
Interpretive Field Studies in Information Systems. MIS Quarterly, 23(1), 67.
https://doi.org/10.2307/249410
231
Kogan, M., & Marina. (2016). Digital Traces of Online Self-Organizing and Problem Solving in
Disaster. In Proceedings of the 19th International Conference on Supporting Group Work -
GROUP ’16 (pp. 479–483). New York, New York, USA: ACM Press.
https://doi.org/10.1145/2957276.2997022
Kogan, M., Palen, L., & Anderson, K. M. (2015). Think Local, Retweet Global. In Proceedings of
the 18th ACM Conference on Computer Supported Cooperative Work & Social Computing -
CSCW ’15 (pp. 981–993). New York, New York, USA: ACM Press.
https://doi.org/10.1145/2675133.2675218
Kong, L., Liu, Z., & Huang, Y. (2014). SPOT. Proceedings of the VLDB Endowment, 7(13), 1681–
1684. https://doi.org/10.14778/2733004.2733060
Krackhardt, D., & Stern, R. N. (1988). Informal Networks and Organizational Crises: An
Experimental Simulation. Social Psychology Quarterly, 51(2), 123.
https://doi.org/10.2307/2786835
Kreps, G. A. (1978). The organization of disaster response: Some fundamental theoretical issues.
In E. L. Quarantelli (Ed.), Disasters: Theory and research (pp. 65–87). London: Sage.
Kropczynski, J., Grace, R., Coche, J., Obeysekare, E., Bénaben, F., Halse, S., … Tapia, A. (2018).
Identifying Actionable Information on Social Media for Emergency Dispatch. In K. Stock &
D. Bunker (Eds.), Proceedings of ISCRAM Asia Pacific 2018 (pp. 1–11). Wellington, New
Zealand. Retrieved from https://prioritydispatch.net/discover_proqa/
Kuijer, L., Jong, A. de, & Eijk, D. van. (2008). Practices as a unit of design. ACM Transactions on
Computer-Human Interaction, 20(4), 1–22. https://doi.org/10.1145/2493382
Kurvinen, E., Koskinen, I., & Battarbee, K. (2008). Prototyping Social Interaction. Design Issues,
24(3), 46–57. https://doi.org/10.1162/desi.2008.24.3.46
Kuutti, K., & Bannon, L. (2014). The turn to practice in HCI. In Proceedings of the 32nd annual
ACM conference on Human factors in computing systems - CHI ’14 (pp. 3543–3552). New
York, New York, USA: ACM Press. https://doi.org/10.1145/2556288.2557111
Kvale, S. (2008). Doing Interviews. Thousand Oaks, CA: Sage. Retrieved from
https://books.google.com/books?hl=en&lr=&id=gkBdBAAAQBAJ&oi=fnd&pg=PR5&dq=
kvale+interviewing&ots=bfzjEMgL6-&sig=07v9t8jI-
WqNFADfiHzAsQl_LuI#v=onepage&q=kvale interviewing&f=false
Lachlan, K. A., Spence, P. R., Lin, X., Najarian, K., & Del Greco, M. (2016). Social media and
crisis management: CERC, search strategies, and Twitter content. Computers in Human
Behavior, 54, 647–652. https://doi.org/10.1016/J.CHB.2015.05.027
LaLone, N., Tapia, A., Zobel, C., Caraega, C., Neppalli, V. K., & Halse, S. (2017). Embracing
human noise as resilience indicator: twitter as power grid correlate. Sustainable and Resilient
Infrastructure, 1–10. https://doi.org/10.1080/23789689.2017.1328920
232
Landwehr, P. M., & Carley, K. M. (2014). Social Media in Disaster Relief (pp. 225–257). Springer,
Berlin, Heidelberg. https://doi.org/10.1007/978-3-642-40837-3_7
Lee, A. S., & Baskerville, R. L. (2003). Generalizing Generalizability in Information Systems
Research. Information Systems Research, 14(3), 221–243.
https://doi.org/10.1287/isre.14.3.221.16560
Leetaru, K., Wang, S., Padmanabhan, A., & Shook, E. (2013). Mapping the global Twitter
heartbeat: The geography of Twitter. First Monday, 18(5).
https://doi.org/10.5210/fm.v18i5.4366
Levine, E. S., & Tisch, J. S. (2014). Analytics in Action at the New York City Police Department’s
Counterterrorism Bureau. Military Operations Research, 19(4), 5–14.
https://doi.org/10.2307/24838523
Li, R., Wang, S., Deng, H., Wang, R., & Chang, K. C.-C. (2012). Towards social user profiling. In
Proceedings of the 18th ACM SIGKDD international conference on Knowledge discovery
and data mining - KDD ’12 (p. 1023). New York, New York, USA: ACM Press.
https://doi.org/10.1145/2339530.2339692
Lin, J., & Cromley, R. G. (2015). Evaluating geo-located Twitter data as a control layer for areal
interpolation of population. Applied Geography, 58, 41–47.
https://doi.org/10.1016/J.APGEOG.2015.01.006
Lin, X., Spence, P. R., Sellnow, T. L., & Lachlan, K. A. (2016). Crisis communication, learning
and responding: Best practices in social media. Computers in Human Behavior, 65, 601–605.
https://doi.org/10.1016/J.CHB.2016.05.080
Liu, Y., Liu, X., Gao, S., Gong, L., Kang, C., Zhi, Y., … Shi, L. (2015). Social Sensing: A New
Approach to Understanding Our Socioeconomic Environments. Annals of the Association of
American Geographers, 105(3), 512–530. https://doi.org/10.1080/00045608.2015.1018773
Ludwig, T., Kotthaus, C., Reuter, C., Dongen, S. van, & Pipek, V. (2017). Situated crowdsourcing
during disasters: Managing the tasks of spontaneous volunteers through public displays.
International Journal of Human-Computer Studies, 102, 103–121.
https://doi.org/10.1016/J.IJHCS.2016.09.008
Malik, M. M., Lamba, H., Nakos, C., & Pfeffer, U. (2015). Population Bias in Geotagged Tweets.
Retrieved from
https://www.aaai.org/ocs/index.php/ICWSM/ICWSM15/paper/viewFile/10662/10551
Manyena, S. B. (2006). The concept of resilience revisited. Disasters, 30(4), 434–450.
https://doi.org/10.1111/j.0361-3666.2006.00331.x
Mark, G., & Semaan, B. (2008). Resilience in collaboration. In Proceedings of the ACM 2008
conference on Computer supported cooperative work - CSCW ’08 (p. 137). New York, New
York, USA: ACM Press. https://doi.org/10.1145/1460563.1460585
233
Martín, Y., Li, Z., & Cutter, S. (2017). Leveraging Twitter to gauge evacuation compliance:
Spatiotemporal analysis of Hurricane Matthew. PLoS One, 12(7).
Martini, S. (2018). The critical role of the telecommunicator in incident command and emergency
support coordination. Public Safety Communications, 28–33. Retrieved from
https://www.apcointl.org/download/cde-47928-role-of-telecommunicator-in-incident-
command/
Marwick, A. E., & boyd, danah. (2011). I tweet honestly, I tweet passionately: Twitter users,
context collapse, and the imagined audience. New Media & Society, 13(1), 114–133.
https://doi.org/10.1177/1461444810365313
Mazer, J. P., Thompson, B., Cherry, J., Russell, M., Payne, H. J., Gail Kirby, E., & Pfohl, W.
(2015). Communication in the face of a school crisis: Examining the volume and content of
social media mentions during active shooter incidents. Computers in Human Behavior, 53,
238–248. https://doi.org/10.1016/J.CHB.2015.06.040
McCormick, S. (2016). New tools for emergency managers: an assessment of obstacles to use and
implementation. Disasters, 40(2), 207–225. https://doi.org/10.1111/disa.12141
McGee, J., Caverlee, J., & Cheng, Z. (2013). Location prediction in social media based on tie
strength. In Proceedings of the 22nd ACM international conference on Conference on
information & knowledge management - CIKM ’13 (pp. 459–468). New York, New York,
USA: ACM Press. https://doi.org/10.1145/2505515.2505544
Mcknight, J., & Kretzmann, J. (1997). Mapping Community Capacity. In M. Minkler (Ed.),
Community organizing and community building for health (pp. 154–174). New Brunswick,
NJ: Rutgers University Press.
McMaster, R., Baber, C., & Duffy, T. (2012). The role of artefacts in police emergency response
sensemaking. In L. Rothkrantz & Z. Franco (Eds.), Proceedings of the 9th International
ISCRAM Conference (pp. 1–10). Vancouver. Retrieved from
http://idl.iscram.org/files/mcmaster/2012/168_McMaster_etal2012.pdf
Medler, B., & Magerko, B. (2010). The implications of improvisational acting and role-playing on
design methodologies. In Proceedings of the 28th international conference on Human factors
in computing systems - CHI ’10 (p. 483). New York, New York, USA: ACM Press.
https://doi.org/10.1145/1753326.1753398
Mergel, I., & Bretschneider, S. I. (2013). A Three-Stage Adoption Process for Social Media Use in
Government. Public Administration Review, 73(3), 390–400.
https://doi.org/10.1111/puar.12021
Middleton, S. E., Middleton, L., & Modafferi, S. (2014). Real-Time Crisis Mapping of Natural
Disasters Using Social Media. IEEE Intelligent Systems, 29(2), 9–17.
https://doi.org/10.1109/MIS.2013.126
234
Miles, M. B., & Huberman, A. M. (1984). Qualitative data analysis: A sourcebook of new methods.
Beverly Hills: Sage. Retrieved from http://bases.bireme.br/cgi-
bin/wxislind.exe/iah/online/?IsisScript=iah/iah.xis&src=google&base=PAHO&lang=p&nex
tAction=lnk&exprSearch=19555&indexSearch=ID
Mitchell, A., Barthel, M., Shearer, E., & Gottfried, J. (2015). The Evolving Role of News on Twitter
and Facebook. Retrieved from http://www.journalism.org/2015/07/14/the-evolving-role-of-
news-on-twitter-and-facebook/
Morstatter, F., Lubold, N., Pon-Barry, H., Pfeffer, J., & Liu, H. (2014). Finding Eyewitness Tweets
During Crises. Retrieved from http://arxiv.org/abs/1403.1773
Morstatter, F., Pfeffer, J., Liu, H., & Carley, K. M. (2013). Is the Sample Good Enough? Comparing
Data from Twitter’s Streaming API with Twitter’s Firehose, 400–408.
https://doi.org/10.1007/978-3-319-05579-4_10
Mulder, F., Ferguson, J., Groenewegen, P., Boersma, K., & Wolbers, J. (2016). Questioning Big
Data: Crowdsourcing crisis data towards an inclusive humanitarian response. Big Data &
Society, 3(2), 205395171666205. https://doi.org/10.1177/2053951716662054
Muralidharan, S., Rasmussen, L., Patterson, D., & Shin, J.-H. (2011). Hope for Haiti: An analysis
of Facebook and Twitter usage during the earthquake relief efforts. Public Relations Review,
37(2), 175–177. https://doi.org/10.1016/J.PUBREV.2011.01.010
Murphy, B. L. (2007). Locating social capital in resilient community-level emergency
management. Natural Hazards, 41(2), 297–315. https://doi.org/10.1007/s11069-006-9037-6
Murthy, D., & Gross, A. J. (2017). Social media processes in disasters: Implications of emergent
technology use. Social Science Research, 63, 356–370.
https://doi.org/10.1016/J.SSRESEARCH.2016.09.015
National Institute of Standards and Technology. (2015). Community Resilience Planning Guide for
Buildings and Infrastructure Systems. NIST Special Publication 1190.
https://doi.org/10.6028/NIST.SP.1190v1
NENA. (2017). 9-1-1 Statistics. Retrieved February 1, 2019, from
https://www.nena.org/page/911Statistics
NENA. (2018a). 9-1-1 Origin & History. Retrieved January 22, 2019, from
https://www.nena.org/page/911overviewfacts
NENA. (2018b). SMS Text-to-9-1-1 Resources for PSAPs & 9-1-1 Authorities. Retrieved January
22, 2019, from https://www.nena.org/page/textresources
NHTSA. (2018). DOT NG911 Initiative Historical Resources. Retrieved January 22, 2019, from
https://www.911.gov/project_dotng911initiativehistoricalresources.html
Nimsgern, C. (2018). Role of the Communications Specialist in ICS: Classification and the
235
Effective Transitioning of Operations. Retrieved January 17, 2019, from
https://www.psconnect.org/blogs/craig-nimsgern/2018/04/21/role-of-the-communications-
specialist-in-ics-class
Norman, D. (2005). Human-centered design considered harmful. Interactions, 12(4), 14–19.
Retrieved from https://interactions.acm.org/archive/view/july-august-2005/human-centered-
design-considered-harmful1
Norri-Sederholm, T., Seppälä, J., Saranto, K., & Paakkonen, H. (2016). Information flow and
situational awareness in emergency medical dispatch. Int. J. Networking and Virtual
Organisations, 16(1), 72–85. Retrieved from
https://pdfs.semanticscholar.org/2cc6/200e278e41347836949ba1aba4defac1cbb7.pdf
Ntuen, C. A., Park, E. H., & Gwang-Myung, K. (2010). Designing an Information Visualization
Tool for Sensemaking. International Journal of Human-Computer Interaction, 26(2–3), 189–
205. https://doi.org/10.1080/10447310903498825
NWS. (2017). Tornado Watch 185. NOAA. Retrieved from
http://www.spc.noaa.gov/products/watch/ww0185.html
Odom, W., Zimmerman, J., Davidoff, S., Forlizzi, J., Dey, A. K., & Lee, M. K. (2012). A fieldwork
of the future with user enactments. In Proceedings of the Designing Interactive Systems
Conference on - DIS ’12 (p. 338). New York, New York, USA: ACM Press.
https://doi.org/10.1145/2317956.2318008
Odom, W., Zimmerman, J., Forlizzi, J., Choi, H., Meier, S., & Park, A. (2014). Unpacking the
thinking and making behind a user enactments project. In Proceedings of the 2014 conference
on Designing interactive systems - DIS ’14 (pp. 513–522). New York, New York, USA: ACM
Press. https://doi.org/10.1145/2598510.2602960
Olteanu, A., Castillo, C., Diaz, F., & Vieweg, S. (2014). CrisisLex: A Lexicon for Collecting and
Filtering Microblogged Communications in Crises. Proc. of the 8th International Conference
on Weblogs and Social Media, 376. https://doi.org/10.1.1.452.7691
Olteanu, A., Vieweg, S., & Castillo, C. (2015). What to Expect When the Unexpected Happens. In
Proceedings of the 18th ACM Conference on Computer Supported Cooperative Work &
Social Computing - CSCW ’15 (pp. 994–1009). New York, New York, USA: ACM Press.
https://doi.org/10.1145/2675133.2675242
Ospina, A. V. (2014). Analysing urban community informatics from a resilience perspective. The
Journal of Community Informatics, 11(1). Retrieved from http://ci-
journal.org/index.php/ciej/article/view/1108
Palen, L., & Anderson, K. M. (2016). Crisis informatics: New data for extraordinary times. Science,
353(6296), 224–225. https://doi.org/10.1126/science.aag2579
Palen, L., Anderson, K. M., Mark, G., Martin, J., Sicker, D., Palmer, M., & Grunwald, D. (2010).
A vision for technology-mediated support for public participation & assistance in mass
236
emergencies & disasters. In Proceedings of the 2010 ACM-BCS Visions of Computer Science
Conference (pp. 1–12). British Computer Society. Retrieved from
http://dl.acm.org/citation.cfm?id=1811194
Palen, L., & Liu, S. B. (2007). Citizen communications in crisis. In Proceedings of the SIGCHI
conference on Human factors in computing systems - CHI ’07 (p. 727). New York, New
York, USA: ACM Press. https://doi.org/10.1145/1240624.1240736
Palen, L., Vieweg, S., Liu, S. B., & Hughes, A. L. (2009). Crisis in a Networked World: Features
of Computer-Mediated Communication in the April 16, 2007, Virginia Tech Event. Social
Science Computer Review, 27(4), 467–480. https://doi.org/10.1177/0894439309332302
Pariser, E. (2011). The filter bubble: What the internet is hiding from you. Penguin.
PEMA. (2009). Volume 1: Basic Plan. Retrieved from
https://www.pema.pa.gov/planningandpreparedness/communityandstateplanning/Pages/Cou
nty-Emergency-Operations-Plan-Toolkit.aspx
Pennsylvania Turnpike Commission. (2018). Operations and Incident Management. Retrieved
December 17, 2018, from https://www.paturnpike.com/travel/emerincident.aspx
Petersen, L., Fallou, L., Reilly, P., & Serafinelli, E. (2018). Public expectations of critical
infrastructure operators in times of crisis. Sustainable and Resilient Infrastructure, 1–16.
https://doi.org/10.1080/23789689.2018.1469358
Plotnick, L., & Hiltz, S. R. (2018). Software Innovations to Support the Use of Social Media by
Emergency Managers. International Journal of Human–Computer Interaction, 34(4), 367–
381. https://doi.org/10.1080/10447318.2018.1427825
Plotnick, L., Hiltz, S. R., Kushma, J. A., & Tapia, A. (2015). Red Tape: Attitudes and Issues Related
to Use of Social Media by U.S. County- Level Emergency Managers. In Proceedings of the
ISCRAM 2015 Conference (pp. 182–192). Retrieved from
http://idl.iscram.org/files/lindaplotnick/2015/1225_LindaPlotnick_etal2015.pdf
Preusse, K. C., & Gipson, C. (2016). Dispatching Information in 911 Teams. Proceedings of the
Human Factors and Ergonomics Society Annual Meeting, 60(1), 117–121.
https://doi.org/10.1177/1541931213601027
Purohit, H., Castillo, C., Imran, M., & Pandey, R. (2018). Social-EOC: Serviceability Model to
Rank Social Media Requests for Emergency Operation Centers. In 2018 IEEE/ACM
International Conference on Advances in Social Networks Analysis and Mining (ASONAM)
(pp. 119–126). IEEE. https://doi.org/10.1109/ASONAM.2018.8508709
Purohit, H., Hampton, A., Bhatt, S., Shalin, V. L., Sheth, A. P., & Flach, J. M. (2014). Identifying
Seekers and Suppliers in Social Media Communities to Support Crisis Coordination.
Computer Supported Cooperative Work (CSCW), 23(4–6), 513–545.
https://doi.org/10.1007/s10606-014-9209-y
237
Quarantelli, E. L. (1984). Emergent Behavior at the Emergency Time Periods of Disasters.
Retrieved from https://apps.dtic.mil/docs/citations/ADA137606
Quarantelli, E. L. (2000). Emergencies, Disasters and Catastrophes Are Different Phenomena.
Disaster Research Center. Retrieved from http://udspace.udel.edu/handle/19716/674
Reed, J. H., Krizman, K. J., Woerner, B. D., & Rappaport, T. S. (1998). An overview of the
challenges and progress in meeting the E-911 requirement for location service. IEEE
Communications Magazine, 36(4), 30–37. https://doi.org/10.1109/35.667410
Reuter, C., Heger, O., & Pipek, V. (2013). Combining Real and Virtual Volunteers through Social
Media. In T. Comes, F. Fiedrich, F. Fortier, J. Geldermann, & T. Müller (Eds.), Proceedings
of the 10th International ISCRAM Conference (pp. 780–790). Baden-Baden. Retrieved from
http://visitmix.com/
Reuter, C., Hughes, A. L., & Kaufhold, M.-A. (2018). Social Media in Crisis Management: An
Evaluation and Analysis of Crisis Informatics Research. International Journal of Human–
Computer Interaction, 34(4), 280–294. https://doi.org/10.1080/10447318.2018.1427832
Reuter, C., & Kaufhold, M.-A. (2018). Fifteen years of social media in emergencies: A
retrospective review and future directions for crisis Informatics. Journal of Contingencies and
Crisis Management, 26(1), 41–57. https://doi.org/10.1111/1468-5973.12196
Reuter, C., Ludwig, T., Friberg, T., & Pratzler-Wanczura, S. (2015). Social Media and Emergency
Services? Interview Study on Current and Potential Use in 7 European Countries.
International Journal of Information Systems for Crisis Response and Management, 7(2), 36–
58. https://doi.org/10.4018/IJISCRAM.2015040103
Reuter, C., Ludwig, T., Kaufhold, M.-A., & Spielhofer, T. (2016). Emergency services׳ attitudes
towards social media: A quantitative and qualitative survey across Europe. International
Journal of Human-Computer Studies, 95(C), 96–111.
https://doi.org/10.1016/j.ijhcs.2016.03.005
Reuter, C., & Spielhofer, T. (2017). Towards social resilience: A quantitative and qualitative survey
on citizens’ perception of social media in emergencies in Europe. Technological Forecasting
and Social Change, 121, 168–180. https://doi.org/10.1016/J.TECHFORE.2016.07.038
Rice, R. G., & Spence, P. R. (2016). Thor visits Lexington: Exploration of the knowledge-sharing
gap and risk management learning in social media during multiple winter storms. Computers
in Human Behavior, 65, 612–618. https://doi.org/10.1016/J.CHB.2016.05.088
Robinson, J. J., Maddock, J., & Starbird, K. (2015). Examining the Role of Human and Technical
Infrastructure during Emergency Response. In Proceedings of the ISCRAM 2015 Conference
(pp. 1–10). Retrieved from
http://faculty.washington.edu/kstarbi/Robinson_Maddock_Starbird_ISCRAM2015.pdf
Rosson, M. B., & Carroll, J. M. (2009). Scenario-based design. In A. Sears & J. Jacko (Eds.),
Human-Computer Interaction: Development Process (pp. 145–164). Boca Raton: CRC Press.
238
Rudra, K., Banerjee, S., Ganguly, N., Goyal, P., Imran, M., & Mitra, P. (2016). Summarizing
Situational Tweets in Crisis Scenario. In Proceedings of the 27th ACM Conference on
Hypertext and Social Media - HT ’16 (pp. 137–147). New York, New York, USA: ACM
Press. https://doi.org/10.1145/2914586.2914600
Saleem, H. M., Xu, Y., & Ruths, D. (2014). Novel Situational Information in Mass Emergencies:
What does Twitter Provide? Procedia Engineering, 78, 155–164.
https://doi.org/10.1016/J.PROENG.2014.07.052
Samuels, R., Taylor, J., & Mohammadi, N. (2018). The Sound of Silence: Exploring How
Decreases in Tweets Contribute to Local Crisis Identification. In K. Boersma & B.
Tomaszewski (Eds.), Proceedings of the 15th ISCRAM Conference (pp. 1–9). Rochester, NY.
Retrieved from
http://idl.iscram.org/files/rachelsamuels/2018/1591_RachelSamuels_etal2018.pdf
Schafer, W. A., Carroll, J. M., Haynes, S. R., & Abrams, S. (2008). Emergency Management
Planning as Collaborative Community Work. Journal of Homeland Security and Emergency
Management, 5(1). https://doi.org/10.2202/1547-7355.1396
Schmidt, K. (1994). Computational mechanisms of interaction: Requirements for a general
notation. COMIC: A Notation for Computational Mechanisms of Interaction.
Schmidt, K. (2008). Cooperative Work and Coordinative Practices (pp. 3–27). Springer, London.
https://doi.org/10.1007/978-1-84800-068-1_1
Schmidt, K., & Bannon, L. (1992). Taking CSCW seriously. Computer Supported Cooperative
Work (CSCW), 1(1–2), 7–40. https://doi.org/10.1007/BF00752449
Schmidt, K., & Simonee, C. (1996). Coordination mechanisms: Towards a conceptual foundation
of CSCW systems design. Computer Supported Cooperative Work (CSCW), 5(2–3), 155–200.
https://doi.org/10.1007/BF00133655
Seale, C. (1999). Quality in Qualitative Research. Qualitative Inquiry, 5(4), 465–478.
https://doi.org/10.1177/107780049900500402
Semaan, B., & Hemsley, J. (2015). Maintaining and Creating Social Infrastructures: Towards a
Theory of Resilience. In L. Palen, M. Büscher, T. Comes, & A. Hughes (Eds.), Proceedings
of the ISCRAM 2015 Conference (pp. 321–328). Kristiansand, Norway. Retrieved from
http://idl.iscram.org/files/bryansemaan/2015/1254_BryanSemaan+JeffHemsley2015.pdf
Semaan, B., & Mark, G. (2011). Technology-mediated social arrangements to resolve breakdowns
in infrastructure during ongoing disruption. ACM Transactions on Computer-Human
Interaction, 18(4), 1–21. https://doi.org/10.1145/2063231.2063235
Shan, Y., Plotnick, L., Hiltz, S., & Yang, L. (2017). A Comparison of Emergency Management
Social Media Use in the United States and England. AMCIS 2017 Proceedings. Retrieved
from http://aisel.aisnet.org/amcis2017/eGovernment/Presentations/1
239
Shelton, T., Poorthuis, A., Graham, M., & Zook, M. (2014). Mapping the data shadows of
Hurricane Sandy: Uncovering the sociospatial dimensions of ‘big data.’ Geoforum, 52, 167–
179. https://doi.org/10.1016/J.GEOFORUM.2014.01.006
Shklovski, I., Palen, L., & Sutton, J. (2008). Finding community through information and
communication technology in disaster response. In Proceedings of the ACM 2008 conference
on Computer supported cooperative work - CSCW ’08 (p. 127). New York, New York, USA:
ACM Press. https://doi.org/10.1145/1460563.1460584
Simsarian, K. (2003). Take it to the next stage. In CHI ’03 extended abstracts on Human factors in
computing systems - CHI ’03 (p. 1012). New York, New York, USA: ACM Press.
https://doi.org/10.1145/765891.766123
Sirkin, D., & Ju, W. (2015). Embodied Design Improvisation: A Method to Make Tacit Design
Knowledge Explicit and Usable (pp. 195–209). Springer, Cham. https://doi.org/10.1007/978-
3-319-06823-7_11
Smith, C. (2018, March). The Controversial Crime-Fighting Program That Changed Big-City
Policing Forever. New York Magazine. Retrieved from
http://nymag.com/intelligencer/2018/03/the-crime-fighting-program-that-changed-new-
york-forever.html
Smith, J., & Osborne, M. (2008). Interpretative phenomenological analysis. In J. Smith (Ed.),
Qualitative Psychology: A Practical Guide to Research Methods (pp. 53–80). London: Sage.
Soden, R., & Palen, L. (2016). Infrastructure in the Wild. In Proceedings of the 2016 CHI
Conference on Human Factors in Computing Systems - CHI ’16 (pp. 2796–2807). New York,
New York, USA: ACM Press. https://doi.org/10.1145/2858036.2858545
St Denis, L. A., Hughes, A. L., & Palen, L. (2012). Trial by Fire: The Deployment of Trusted
Digital Volunteers in the 2011 Shadow Lake Fire. In Proceedings of the 9th International
ISCRAM Conference (pp. 1–10). Vancouver.
Stallings, R. A., & Quarantelli, E. L. (1985). Emergent Citizen Groups and Emergency
Management. Public Administration Review, 45, 93. https://doi.org/10.2307/3135003
Star, S. L., & Ruhleder, K. (1996). Steps Toward an Ecology of Infrastructure: Design and Access
for Large Information Spaces. Information Systems Research, 7(1), 111–134.
https://doi.org/10.1287/isre.7.1.111
Starbird, K., Dailey, D., Walker, A. H., Leschine, T. M., Pavia, R., & Bostrom, A. (2015). Social
Media, Public Participation, and the 2010 BP Deepwater Horizon Oil Spill. Human and
Ecological Risk Assessment: An International Journal, 21(3), 605–630.
https://doi.org/10.1080/10807039.2014.947866
Starbird, K., & Kate. (2013). Delivering patients to sacré coeur. In Proceedings of the SIGCHI
Conference on Human Factors in Computing Systems - CHI ’13 (p. 801). New York, New
240
York, USA: ACM Press. https://doi.org/10.1145/2470654.2470769
Starbird, K., Muzny, G., & Palen, L. (2012). Learning from the Crowd: Collaborative Filtering
Techniques for Identifying On-the-Ground Twitterers during Mass Disruptions. In
Proceedings of the 9th International ISCRAM Conference. Retrieved from
http://www.iscramlive.org/ISCRAM2012/proceedings/148.pdf
Starbird, K., & Palen, L. (2011). "Voluntweeters" In Proceedings of the 2011 annual
conference on Human factors in computing systems - CHI ’11 (p. 1071). New York, New
York, USA: ACM Press. https://doi.org/10.1145/1978942.1979102
Starbird, K., & Palen, L. (2012). (How) will the revolution be retweeted? In Proceedings of the
ACM 2012 conference on Computer Supported Cooperative Work - CSCW ’12 (p. 7). New
York, New York, USA: ACM Press. https://doi.org/10.1145/2145204.2145212
Starbird, K., & Palen, L. (2013). Working and sustaining the virtual "Disaster Desk"
In Proceedings of the 2013 conference on Computer supported cooperative work - CSCW ’13
(p. 491). New York, New York, USA: ACM Press. https://doi.org/10.1145/2441776.2441832
Starbird, K., Palen, L., Hughes, A. L., & Vieweg, S. (2010). Chatter on the red what hazards threat
reveals about the social life of microblogged information. In Proceedings of the 2010 ACM
conference on Computer supported cooperative work - CSCW ’10 (p. 241). New York, New
York, USA: ACM Press. https://doi.org/10.1145/1718918.1718965
Stasko, J., Görg, C., & Liu, Z. (2008). Jigsaw: Supporting Investigative Analysis through
Interactive Visualization. Information Visualization, 7(2), 118–132.
https://doi.org/10.1057/palgrave.ivs.9500180
Stern. (2007). On Solid Ground: Essential Properties for Growing Grounded Theory. In A. Bryant
& K. Charmaz (Eds.), The SAGE Handbook of Grounded Theory (pp. 114–127). London:
Sage.
Strauss, A. (1985). Work and the Division of Labor. The Sociological Quarterly, 26(1), 1–19.
https://doi.org/10.1111/j.1533-8525.1985.tb00212.x
Strauss, A., & Corbin, J. (1990). Basics of qualitative research: Grounded theory procedures and
techniques. Thousand Oaks, CA: Sage. Retrieved from http://psycnet.apa.org/record/1990-
98829-000
Strauss, A. L. (1987). Qualitative analysis for social scientists. Cambridge University Press.
Sutton, J., Gibson, C. Ben, Phillips, N. E., Spiro, E. S., League, C., Johnson, B., … Butts, C. T.
(2015). A cross-hazard analysis of terse message retransmission on Twitter. Proceedings of
the National Academy of Sciences of the United States of America, 112(48), 14793–14798.
https://doi.org/10.1073/pnas.1508916112
Sutton, J., Palen, L., & Shklovski, I. (2008). Emergent Uses of Social Media in the California
Wildfires Backchannels on the Front Lines: Emergent Uses of Social Media in the 2007
241
Southern California Wildfires. In Proceedings of the 5th International ISCRAM Conference
(pp. 624–632). Washington D.C. Retrieved from
http://www.iscram.org/legacy/dmdocuments/ISCRAM2008/papers/ISCRAM2008_Sutton_e
tal.pdf
Sutton, J., Spiro, E. S., Johnson, B., Fitzhugh, S., Gibson, B., & Butts, C. T. (2014). Warning
tweets: serial transmission of messages during the warning phase of a disaster event.
Information, Communication & Society, 17(6), 765–787.
https://doi.org/10.1080/1369118X.2013.862561
Svanaes, D., & Seland, G. (2004). Putting the users center stage. In Proceedings of the 2004
conference on Human factors in computing systems - CHI ’04 (pp. 479–486). New York,
New York, USA: ACM Press. https://doi.org/10.1145/985692.985753
Tagliacozzo, S., & Arcidiacono, C. (2015). Do ICTs Help To Maintain Social Capital In The
Disaster Recovery Phase? A Case Study Of The L’aquila Earthquake. The Journal of
Community Informatics, 11(1). Retrieved from http://ci-
journal.org/index.php/ciej/article/view/1085
Takahashi, B., Tandoc, E. C., & Carmichael, C. (2015). Communicating on Twitter during a
disaster: An analysis of tweets during Typhoon Haiyan in the Philippines. Computers in
Human Behavior, 50, 392–398. https://doi.org/10.1016/J.CHB.2015.04.020
Tapia, A. H., Bajpai, K., Jansen, B. J., & Yen, J. (2011). Seeking the Trustworthy Tweet: Can
Microblogged Data Fit the Information Needs of Disaster Response and Humanitarian Relief
Organizations. In Proceedings of the 8th International ISCRAM Conference (pp. 1–10).
Lisbon, Portugal. Retrieved from
http://idl.iscram.org/files/tapia/2011/991_Tapia_etal2011.pdf
Tapia, A. H., & Moore, K. (2014). Good Enough is Good Enough: Overcoming Disaster Response
Organizations’ Slow Social Media Data Adoption. Computer Supported Cooperative Work
(CSCW), 23(4–6), 483–512. https://doi.org/10.1007/s10606-014-9206-1
Tapia, A. H., Moore, K. A., & Johnson, N. J. (2013). Beyond the Trustworthy Tweet: A Deeper
Understanding of Microblogged Data Use by Disaster Response and Humanitarian Relief
Organizations. In Proceedings of the 10th International ISCRAM Conference (pp. 770–779).
Baden-Baden, Germany. Retrieved from
http://help.iscram.org/legacy/ISCRAM2013/files/121.pdf
Tien, I., Aibek, M., Benas, D., & C. Pu. (2016). Detection of damage and failure events of critical
public infrastructure using social sensor big data. In Proceedings of International Conference
on Internet of Things and Big Data.
Tim, Y., Pan, S. L., Ractham, P., & Kaewkitipong, L. (2017). Digitally enabled disaster response:
the emergence of social media as boundary objects in a flooding disaster. Information Systems
Journal, 27(2), 197–232. https://doi.org/10.1111/isj.12114
Toepke, S. L. (2018). Leveraging Elasticsearch and Botometer to Explore Volunteered Geographic
242
Information. In K. Boersma & B. Tomaszewski (Eds.), Proceedings of the 15th ISCRAM
Conference (pp. 663–676). Rochester, NY. Retrieved from
http://idl.iscram.org/files/samuelleetoepke/2018/1588_SamuelLeeToepke2018.pdf
Trauth, E. M., & Jessup, L. M. (2000). Understanding Computer-Mediated Discussions: Positivist
and Interpretive Analyses of Group Support System Use. MIS Quarterly, 24(1), 43.
https://doi.org/10.2307/3250979
Trauth, E. M., Quesenberry, J. L., & Huang, H. (2009). Retaining women in the U.S. IT workforce:
theorizing the influence of organizational factors. European Journal of Information Systems,
18(5), 476–497. https://doi.org/10.1057/ejis.2009.31
Twitter. (2016). Streaming API request parameters: Locations. Retrieved from
https://dev.twitter.com/streaming/overview/request-parameters#locations
Valkonen, P., & Liinasuo, M. (2010). Role playing with fire fighters. In Proceedings of the 6th
Nordic Conference on Human-Computer Interaction Extending Boundaries - NordiCHI ’10
(p. 805). New York, New York, USA: ACM Press. https://doi.org/10.1145/1868914.1869034
Veil, S. R., Buehner, T., & Palenchar, M. J. (2011). A Work-In-Process Literature Review:
Incorporating Social Media in Risk and Crisis Communication. Journal of Contingencies and
Crisis Management, 19(2), 110–122. https://doi.org/10.1111/j.1468-5973.2011.00639.x
Vieweg, S., Castillo, C., & Imran, M. (2014). Integrating Social Media Communications into the
Rapid Assessment of Sudden Onset Disasters (pp. 444–461). Springer, Cham.
https://doi.org/10.1007/978-3-319-13734-6_32
Vieweg, S., Hughes, A. L., Starbird, K., & Palen, L. (2010). Microblogging during two natural
hazards events: what twitter may contribute to situational awareness. In Proceedings of the
28th international conference on Human factors in computing systems - CHI ’10 (p. 1079).
New York, New York, USA: ACM Press. https://doi.org/10.1145/1753326.1753486
Wang, T.-Y., Harper, F. M., & Hecht, B. (2014). Designing Better Location Fields in User Profiles.
In Proceedings of the 18th International Conference on Supporting Group Work - GROUP
’14 (pp. 73–80). New York, New York, USA: ACM Press.
https://doi.org/10.1145/2660398.2660424
Weick, K. E., Sutcliffe, K. M., & Obstfeld, D. (2005). Organizing and the Process of Sensemaking.
Organization Science, 16(4), 409–421. https://doi.org/10.1287/orsc.1050.0133
White, J. I., Palen, L., & Anderson, K. M. (2014). Digital mobilization in disaster response. In
Proceedings of the 17th ACM conference on Computer supported cooperative work & social
computing - CSCW ’14 (pp. 866–876). New York, New York, USA: ACM Press.
https://doi.org/10.1145/2531602.2531633
Whittaker, S. (2011). Personal information management: From information consumption to
curation. Annual Review of Information Science and Technology, 45(1), 1–62.
https://doi.org/10.1002/aris.2011.1440450108
243
Williamson, K. (2013). Questionnaires, individual interviews and focus group interviews. In K.
Williamson & J. Johanson (Eds.), Research methods: Information, systems, and contexts (pp.
349–372). Pharan, Australia: Tilde University Press.
Winograd, T., Woods, D., Miller, J., Jeffries, R., Fischer, G., Garcia, O., … Mountford, S. (1997).
The challenge of human-centered design (NSF workshop on human-centered systems).
Retrieved from https://www.researchgate.net/publication/261133023
Wolbers, J., & Boersma, K. (2013). The Common Operational Picture as Collective Sensemaking.
Journal of Contingencies and Crisis Management, 21(4), 186–199.
https://doi.org/10.1111/1468-5973.12027
Wukich, C. (2016). Government Social Media Messages across Disaster Phases. Journal of
Contingencies and Crisis Management, 24(4), 230–243. https://doi.org/10.1111/1468-
5973.12119
Wukich, C. (2018). Preparing for Disaster: Social Media Use for Household, Organizational, and
Community Preparedness. Risk, Hazards & Crisis in Public Policy, rhc3.12161.
https://doi.org/10.1002/rhc3.12161
Xu, Z., Luo, X., Liu, Y., Choo, K.-K. R., Sugumaran, V., Yen, N., … Hu, C. (2016). From Latency,
through Outbreak, to Decline: Detecting Different States of Emergency Events Using Web
Resources. IEEE Transactions on Big Data, 1–1.
https://doi.org/10.1109/TBDATA.2016.2599935
Zade, H., Shah, K., Rangarajan, V., Kshirsagar, P., Imran, M., & Starbird, K. (2018). From
Situational Awareness to Actionability. Proceedings of the ACM on Human-Computer
Interaction, 2(CSCW), 1–18. https://doi.org/10.1145/3274464
Zhang, J., Sun, J., Zhang, R., & Zhang, Y. (2015). Your actions tell where you are: Uncovering
Twitter users in a metropolitan area. In 2015 IEEE Conference on Communications and
Network Security (CNS) (pp. 424–432). IEEE. https://doi.org/10.1109/CNS.2015.7346854
Zhang, Y., Drake, W., Li, Y., Zobel, C., & Cowell, M. (2015). Fostering Community Resilience
through Adaptive Learning in a Social Media Age: Municipal Twitter Use in New Jersey
following Hurricane Sandy. In L. Palen, M. Buscher, T. Comes, & A. Hughes (Eds.),
Proceedings of the ISCRAM 2015 Conference (pp. 1–9). Kristiansand, Norway.
Zheng, X., Han, J., & Sun, A. (2017). A Survey of Location Prediction on Twitter. Retrieved from
http://arxiv.org/abs/1705.03172
244
Appendices
Appendix A
Interview Protocol
I BACKGROUND
Can you tell me a little about your professional background? (e.g. How long in job; prior
experiences in public safety, etc.)
Can you tell me about your [public safety organization]?
• How many personnel? What are their roles?
• What is its administrative position within local government?
What types of information do you communicate to the public in your job?
What does situational awareness mean for you as a [job title]?
II EXPERIENCE
Can you tell me about the types of emergencies you typically experience in your job?
Resources & Activities
• How do become aware of emergencies?
• What tools or means do you use to communicate emergencies to the community?
• What type of information related to emergencies interests you locally?
regionally/state? Nationally?
o How do you seek and access this information?
Tasks & Domains
• What officials and organizations do you interact with to respond to emergencies?
• What citizens and community organizations do you cooperate with? How do they
participate in emergency response?
How do you prepare and plan for emergencies that might occur in the community?
Resources & Activities
• How do you plan for emergencies?
• What type of information do you use for emergency planning?
o How do you seek and access this information?
245
Tasks & Domains
• What officials and organizations do you interact with to plan for emergencies?
• What citizens and community organizations do you cooperate with in emergency
planning? How do they contribute?
III SOCIAL MEDIA EXPERIENCE
Do you use social media in your work? If so, in what ways??
• How do you use social media to disseminate information?
• Do you monitor information on social media?
o What kinds of information? How?
What barriers of challenges are involved in the use of social media for information
dissemination? Monitoring?
• With respect to staff, policy, tools, trust?
Can you tell me an example of when social media proved effective in your job? When it
proved ineffective?
• Why was the information useful?
• Why/when is social media information trustworthy (accurate and credible)?
IV SCEANARIO 1
What are your first impressions regarding the plausibility of the scenario described?
Do you have any questions regarding the idea of community volunteers or how they might
use social media to contribute to emergency response and management?
Have you encountered a situation such as described in the scenario?
Did you or can you imagine using social media in the was described?
What about the utility of community volunteers distributing/monitoring information on
social media?
What seems the most and least useful in the scenario described?
V SCENARIO 2
What are your first impressions regarding the plausibility of the scenario described?
Do you have any questions regarding the idea of community volunteers or how they might
use social media to contribute to emergency response and management?
246
Have you encountered a situation such as described in the scenario?
Did you or can you imagine using social media in the was described?
What about the utility of community volunteers distributing/monitoring information on
social media?
What seems the most and least useful in the scenario described?
VI SCENARIO 3
What are your first impressions regarding the plausibility of the scenario described?
Do you have any questions regarding the idea of community volunteers or how they might
use social media to contribute to emergency response and management?
Have you encountered a situation such as described in the scenario?
Did you or can you imagine using social media in the was described?
What about the utility of community volunteers distributing/monitoring information on
social media?
What seems the most and least useful in the scenario described?
VII CONCLUSION
Overall, how might community volunteers assist you in:
• Distributing information to the public?
• Monitoring useful information?
What challenges would be involved in the use of citizen volunteers as described in the
scenarios?
What challenges do you see related to social media use and…
• Current staffing and available time among officials in your agency?
• Technologies for distributing / monitoring information?
• Policies related to such use?
• Trust in information reported on social media?
Thank you for your time.
247
Appendix B
IRB Approval
This research was approved by The Pennsylvania State Institutional Review Board (#5334).
248
Appendix C
Funding
This research was supported by National Science Foundation grants #541155 and #741370.
Rob Grace E320 Westgate, University Park, PA 16802
(540) 460-6190 • [email protected]
• Education
PhD in Informatics Pennsylvania State University, University Park, PA
2019
MS in Information Science and Technology Pennsylvania State University, University Park, PA
2015
BA in English and History, summa cum laude Texas A&M University, College Station, TX
2008
• Research Experience
• Research Assistant College of Information Sciences and Technology, Penn State
2012-2018
• Teaching Experience
• Instructor College of Information Sciences and Technology, Penn State
2017-2018
• • Teaching Assistant
College of Information Sciences and Technology, Penn State
2014-2018
• • English Teacher
Songwon High School, Gwangju, South Korea
2009-2011
• References
Fred Fonseca, Associate Professor of Information Sciences and Technology Penn State University, University Park, PA [email protected] Andrea Tapia, Associate Professor of Information Sciences and Technology Penn State University, University Park, PA [email protected] Jess Kropczynski, Assistant Professor of Information Technology University of Cincinnati, Cincinnati, OH [email protected]