2.1 Background
The bombing of the Boston Marathon on April 15, 2013 resulted in 3 deaths and more than 250 casualties. Over the subsequent week police engaged in an area-wide manhunt that resulted in a ‘lockdown’ of Boston and neighboring suburbs [
18]–[
20]. Citizens of Boston, who were at one point instructed directly by the Governor of Massachusetts to stay in their homes to avoid danger, were understandably afraid. Examples were seen on Twitter: ‘Just can’t shake the horror of yesterday. A search warrant has been executed in a Revere apt complex, even closer to home.’ ‘Holy shit first the bombing now a shooting at MIT can’t people just leave Boston alone.’ Bostonians were not the only ones who became afraid, however. Those with loved ones who lived in Boston, or had traveled to Boston for the marathon, worried for their safety, e.g., ‘Thinking of all my family and people I know in Massachusetts/Boston area and imagining their fear is heartbreaking.’ For others, their sense of the benevolence of the world was shaken [
7], e.g., ‘This world is becoming a truly scary place. As though bombing a marathon wasnt enough.’ ‘Seriously what the hell is going on in this city… #peoplesuck #gunshots #explosions #getoutofhere #bostonstrong.’
Fear was not the only response in Boston, however, or in its fellow communities. The slogan ‘Boston Strong’ emerged as a rallying cry to demonstrate the city’s defiance in response to the terrorists’ attempt to intimidate them [
21]. Boston-area residents organized community events such as the ‘Run to Remember’ memorial foot-race and music concerts. Just as the attacks spread fear well beyond Boston itself, expressions of social support were also directed to Boston from around the globe. In New York, fans of the New York Yankees baseball team, the bitter rival of Boston’s team (the Red Sox), stood in their stadium to sing ‘Sweet Caroline,’ the Red Sox anthem [
22]. In Chicago, more than 200 runners gathered for a run of solidarity [
23]. A week after the Boston Marathon, thousands of marathon runners in London wore a black ribbon in solidarity with the people of Boston [
24].
Interestingly, such extensions of support across community-community relations have been observed in other circumstances when tragic or disastrous events befall a particular locale [
25]. What do these reactions reveal about the nature of social organization in the 21st century? Recent technical and social innovations have enabled individuals to break out from their traditional geographic and cultural boundaries, creating a ‘network society’ [
26]. This de-coupling of the social from the geographic has given individuals increased personal autonomy but has also created challenges for civic institutions concerned with supporting their needs and anticipating their behavior [
27], [
28]. Does London have a particular kind of social tie to Boston? If so, what is its basis? Are these ties based on inter-personal relationships, or on the basis of shared identification with some set of cultural symbols, such as the rituals around baseball that appear to have motivated the responses of Yankee fans [
29], [
30]. What motivates these inter-communal relationships? What characteristics of the relationships between communities might lead one to respond when the other is attacked or befallen by disaster, and how do these inter-communal relationships influence the impact and effectiveness of terrorist acts in promoting fear?
2.2 Ripples of fear and related emotions
According to legal scholar Cass Sunstein [
1] ‘Terrorists show a working knowledge of’ the dynamics of fear, in particular, the kinds of spectacles that prompt the general public to become afraid. Experts typically evaluate threats, such as those posed by terrorist attacks, in terms of expected costs and casualties derived from the probability of an attack and its typical consequences [
5]. By contrast, the public’s responses to threats are subject to what risk researchers call
social amplification[
5], [
31]. In social amplification, individuals heighten their emotion and communication about some threats and attenuate their responses to others in a manner that is inconsistent with expert calculations of their expected outcomes. Threats are socially amplified because individuals appraise and interpret them in the context of personal circumstances, social relationships, and institutional expectations that include factors that go beyond the calculable, physical consequences constituted by the threats themselves. Such contexts include, for example, concerns about the quality of the information they are receiving or worries that their government is not competently addressing the problem [
6].
Terrorist attacks are designed to be amplified in this way [
32]. First, the vivid and spectacular nature of terrorist violence draws increased attention and intense emotion [
6], [
33]. The irrevocable and horrifying nature of the terrorist outcomes also invokes
dread[
6], [
31], [
33], [
34], a powerful feature in the evaluation of risk. When individuals dread an outcome, they are more likely to take action and to demand that others take action. They become insensitive to rational appeals and trade-offs as fear motivates them to insist the outcome be avoided at almost any cost [
1], [
31]. Second, the unexpected nature of the attack also serves what researchers refer to as a
signalling function. An event has high signalling value when its occurrence suggests that previously unconsidered risks deserve greater attention and that existing institutions and authorities are not adequately prepared to deal with a threat [
6], [
34]. In this way, terrorism undermines the public’s confidence in its own government and institutions [
35], [
36]. Furthermore, the fearful individuals are likely to over-estimate the likelihood of future terrorist events [
3], [
12], [
33].
A third feature of terrorist acts is their relationship to groups and group identities. Self-categorization refers to the cognitive processes that lead people to identify with particular groups, such as ethnic, regional, or national communities [
11], [
29]. Social identity refers to the evaluations individuals make of themselves and others based on their group membership [
29]. Disasters tend to impact the ways in which people self-categorize themselves. Individuals normally participate and identify with many distinct (but somewhat overlapping) communities [
29]. When one such group is threatened, however, this identity becomes more salient and many individuals will more closely identify with it [
37]. The effect is to increase solidarity within the ‘in-group,’ with individuals becoming more charitable and supportive of those who they recognize sharing their identity [
38], while simultaneously becoming more mistrustful of and hostile toward those they perceive to be ‘outsiders’ [
37]. These effects are particularly important in the context of terrorist attacks. Unlike natural disasters, terrorist attacks are socio-political acts directed by members of one group against another. Thus, in addition to signalling that members of one’s group are exposed to a previously unanticipated risk, terrorist attacks indicate that these risks are due to the planned, intentional behavior of an out-group [
13], [
39]. For example, after the attacks of 9/11, Americans all over the country responded by prominently displaying American flags [
16]. At the same time, many Americans felt more hostile toward Muslim nations and violence against Muslims increased [
12].
All of these non-objective, social factors lead the amplification of fear and related emotions in response to terrorist attacks to ‘ripple’ out from the geographic locale in which the attack takes place. Risk researchers define ‘ripple’ effects as the effect of an event on emotions, appraisals and decisions in geographically and temporally distant contexts. Ripples can travel over many paths. Vivid images are likely to be carried and shared through media [
40], bringing information about the attack to a larger and geographically dispersed audience. At the same time, the ‘signal’ that existing understandings and institutions have not properly assessed risk and threats leads to the broader conclusion that no one is safe anywhere [
41], [
42]. Group identification also has a potentially large geographic reach. In a conflict between groups, individuals may feel compelled to choose sides or risk being identified as members of the out-group by both communities [
29]. For example, after the 9/11 attacks almost half of Germans surveyed say they agreed with the statement ‘we are all Americans now.’
In addition to these ripples motivated by identifiable processes, there is also evidence that expressions of emotion are themselves contagious. For example, [
9] found that rainfall in one location slightly depresses both residents of that geographic area and their friends at distant locations. Kramer et al. [
10] find similar results using a controlled experiment. These findings suggest that the mere incidence of an emotion, such as fear, at a distant location may encourage others to take on that emotion, even if they do not identify with the group that is attacked or believe that the attack is a signal of a new threat.
These ripple effects, and ripples of fear in particular, have important practical consequences for first responders and other civic authorities who must respond to terrorist attacks. Fear can lead to costly and irrational decision-making, including the undertaking of behaviors with greater objective risk [
1], [
43]. For example, after the attacks of 9/11, many individuals substituted driving for flying in planes, actually increasing personal and public risk [
1]. The inability to address fears through concrete action can also lead to its persistence over time, a condition associated with post-traumatic stress disorder (PTSD) [
7], [
34], [
44]. Fear can also place demands on local infrastructure, such as the need for increased law enforcement or the perception that local or national governments are not competent or addressing the needs of the people [
36], and can even influence long term assessments of the economy [
45]. Fear can lead individuals to lean on unfair, irrational biases that unfairly target others for harm [
46]. In particular, when individuals perceive that the threatening situation has been brought under control, fear is likely to convert to anger which then motivates attacks against outsiders or others perceived to be a source of threat [
13], [
36].
The connection that people feel to those who were attacked can also spur positive emotions and actions. There is evidence that an individual’s emotional proximity to a tragedy is associated with their willingness to extend social support to those in need [
47], and that when individuals self-categorize as a members of the threatened community they perceive it as something that effects ‘us’ rather than a set of others who are distinct from ‘me’ [
48]. For example, after a shooting at Northern Illinois University in February, 2008 students from Virginia Tech University, where there had been a shooting only 10 months before, offered extensive support to their newly victimized ‘neighbors’ [
25]. It is thus possible that fear and support are linked. Those who feel a close connection with the distant, effected community will likely feel the threat more acutely and personally, making them both more afraid and yet more willing to help.
The preceding argument suggests there is substantial theoretical and practical value in understanding where ripples of fear and related emotions are likely to flow. When a community in one geographic location is directly attacked, where else do people become afraid? Such information would be useful to civic authorities seeking to anticipate how their local communities might be effected by distant events. It would also shed theoretical light on the social processes most prominent in the diffusion of fear and other responses as well as new insights into how they operate.
To date, however, little research has focused on these emotion spreading processes in response to real-world terrorist attacks or disaster events more broadly. Most studies of emotional responses to terrorist attacks focus on those in the area directly effected. When individuals from other communities are included in the analysis they are used for the purposes of comparison and control [
36], [
44], [
49] or are treated as a single group characterized as a whole [
7], [
15], [
17]. Conversely, studies of emotional contagion and other forms of diffusion tend to focus on the way that individualized emotions and information about personal events are communicated over social network ties [
9], [
10], [
50]. Thus, these studies provide little insight into how fear or other feelings diffuse when relevant information and emotion is simultaneously broadcast both over social ties and through other channels, such as mass media [
51], [
52].
Perhaps one of the reasons for limited study in this area is the difficulty of observing geographically dispersed behavior in the immediate aftermath of a terrorist attack. However, social media data offer the potential to overcome this limitation.
Social media data have been gaining increasing attention in studies of emergencies and disasters, a field often referred to as
crisis informatics[
53]. Social media data are useful because they contain fine-grained information regarding the timing and location of individual responses. This permits researchers to analyze real-world behaviors in close proximity to dangerous and disastrous events without them having to anticipate where and when they will occur.
Individuals often rely on social media in emergency circumstances to support a variety of information needs that are not fully served by authorities or established news outlets. Social media facilitate ‘backchannel communication’ [
8] by allowing users to obtain timely and accurate information tailored to their local context [
8], [
53]–[
63]. In addition to providing information about the threatening event itself, social media can also help coordinate efforts to provide assistance [
64], [
65].
Consistent with findings from studies of more general behavior, research indicates that individuals who are in areas directly under threat use social media differently than those who are more distant. In particular, directly affected individuals are significantly more likely to share specific, locally relevant information, whereas more distant individuals are more likely to share generic information, such as an image from the disaster [
40]. This intensity of information sharing on social media, particularly among locals, has become useful for detecting events or circumstances before they reach mainstream media. Twitter has been utilized as an early detection system for emerging public health problems [
66]–[
68], real-time emergency detection [
69]–[
71] and crisis management [
65], [
72], [
73].
The focus of these analyses has largely been on gathering information about the physical realities that correspond to the threat, however, rather than the perceptions and emotions that result from it. Thus, researchers have placed emphasis on the spread of rumor and misinformation with the aim of identifying and removing inaccuracies [
65], [
74]–[
76]. Less attention has been paid to the diffusion of emotions and social realities, such as group identification, that also follow from the social amplification of threats. One exception is [
77], who map the expression of positive and negative emotion in response to Hurricane Sandy. Their findings focus on the unique features of those directly effected by the storm, who are more likely to tweet negative sentiments, rather than on a comparison between the unaffected areas.
Such comparisons are the focus of our work, which advances a systematic understanding of the inter-communal relationships that facilitate ripples of emotion by comparing tweets issued from different cities in response to the Boston Marathon bombings to other observable relationships between these cities and Boston. We employ a concept-based affective lexicon SentiSense [
78] to extract different categories of sentiments (e.g., anger, fear, joy, etc.) from the tweets. We then compare these sentiments to three pre-existing ties between each city and Boston: geographic distance, person-to-person relationships, and personal visits between cities. Although other approaches such as LIWC Dictionary or more recent sentiment detection techniques [
79], [
80] are also applicable, SentiSense allows us to efficiently detect multiple emotional categories grounded in psychological theories with broad vocabulary coverage. Using this approach, the scope of our analysis is limited to content written in English. Nevertheless, through the temporal analysis of changes in the sentiment experession captured by a broad set of lexicons, our goal is to develop a systematic understanding regarding how users’ expression of fear varied in response to the Marathon bombings. By analyzing the responses of communities outside of Boston to an attack on Boston we ask:
• Which communities, outside of Boston, shared fear with Boston in the wake of the bombings?
• What factors - geographic, social, or experiential - best explain these different levels of community fear?
• Are these same factors associated with expressions of social support?
• How is fear itself associated with expressions of social support?
We discuss our method and results in the following sections.