The emotional antecedents of solidarity in social media crowds

Abstract

This study examines the organizational dynamics of social media crowds, in particular, the influence of a crowd’s emotional expression on its solidarity. To identify the relationship between emotions expressed and solidarity, marked by sustained participation in the crowd, the study uses tweets from a unique population of crowds—those tweeting about ongoing National Football League games. Observing this population permits the use of game results as quasi-random treatments on crowds, helping to reduce confounding factors. Results indicate that participation in these crowds is self-sustaining in the medium term (1 week) and can be stimulated or suppressed by emotional expression in a short term (1 hour), depending on the discrete emotion expressed. In particular, anger encourages participation while sadness discourages it. Positive emotions and anxiety have a more nuanced relationship with participation.

Keywords

Crowds emotions natural experiment social media solidarity

Social media crowds are becoming increasingly prominent. Such crowds now regularly form around televised political debates (Vaccari et al., 2015), social movements (Jackson and Foucault Welles, 2015; Papacharissi and De Fatima Oliveira, 2012), natural disasters, and terrorist attacks (Lin and Margolin, 2014; Sutton et al., 2015).

The impact of these crowds has garnered considerable research attention. Social media crowds can have substantial influence and power, calling attention to long overlooked grievances (Freelon et al., 2018; Jensen and Bang, 2013) or providing social support to those in need (Lin and Margolin, 2014). Yet, as researchers have recently pointed out (Freelon et al., 2018; Lin, 2015), little research has systematically theorized and examined the organizational dynamics of these crowds.

This article initiates such a theoretical inquiry by examining the role that the expression of discrete emotions, such as anger or sadness, plays in sustaining participation in a crowd. While emotional expression has long been a signature feature of crowds (Lin, 2015), and some researchers have argued that this expression has a collective impact on crowd dynamics (Collins, 2004), we introduce a theoretical mechanism explaining how macro-level crowd behavior can develop through individual-level emotional expressions. In particular, we argue that shared events at the crowd level stimulate individuals to express emotions. When pooled by social media in a common digital space, the aggregate of these emotional expressions re-orients crowd members toward the crowd or their fellow members, finding, for example, the crowd more (or less) attractive or feeling more (or less) committed to it, depending on the distribution of discrete emotions expressed. The result is an organizing dynamic through which shared events restructure the crowd, enhancing or undermining its ability to sustain itself via individuals’ communicative behaviors.

Such models of spontaneous or “self” organization have been proposed for analyzing collective dynamics before (Monge and Contractor, 2003) and increasingly for understanding social movements and collective action in the digital age (Bennett and Segerberg, 2012). However, previous research tends to focus on higher order logics of organizing (Stohl, 2014), such as the explicit ideologies and values articulated in the narratives deployed by crowd participants (Bennett et al., 2011). For example, movements that value open, diverse participation may organize through looser affiliations than those with a more rigid, controlled collective action frames (Bennett and Segerberg, 2011). By contrast, this study focuses on a basic mechanism—emotional expression—that applies to crowds based on their generic structure—the co-presence of individuals sharing some experience and interests.

Identifying such a mechanism presents a methodological challenge, however, because many real-world crowds carry unique goals and values, heterogeneously respond to various events and external stimuli, and often fail to reoccur more than a few times. These factors limit the ability to pool crowds into a common analysis framework, making it difficult to identify the generic mechanisms. To address this challenge, we examine a population of social media crowds—fans tweeting about National Football League (NFL) teams before, during, and after games—where ideologies and values are homogeneous, the characteristics of shared events can be measured, and the crowd gatherings are repeated multiple times. Furthermore, this empirical setting resembles quasi-experiments in which crowds are treated with shared events that crowd members have limited control over, thus helping to eliminating many confounding factors of crowd behaviors, as we further describe in section ‘Methods’.

The rest of this article is organized as follows. We begin with the variable of broadest interest—self-sustained participation in the crowd as a reflection of the crowd’s solidarity. We then examine and test a two-step mechanism through which the expression of discrete emotions in response to a shared event differently influences solidarity in social media crowds.

Crowd solidarity: the self-sustenance of a proto-organization

Crowds were originally defined as spontaneous gatherings, generally responding to shared events (e.g. disturbing or surprising events in relation to the fate or status of a social group, Lin, 2015). Social media crowds are similar, with the requirement of geographic proximity relaxed. Lin (2015) thus defines social media crowds as “a large number of people who gather together in the same space at the same time, where the space is an online social media platform” (p. 235).

Crowds constitute a form of proto-organization, that is, an early stage in the spontaneous organization of collective action (Jensen and Bang, 2013; Stohl, 2014). They can assemble substantially through individual motivations, such as to express personal opinions or simply to be entertained by a shared event. However, once assembled in a common space, these individually motivated participants are susceptible to collective dynamics, such as shared attention or emotional contagion that can spur further organization and collective action (Lin, Keegan, Margolin, & Lazer, 2014).

The extent to which a crowd can spur larger action depends substantially on its ability to retain the interest, engagement, and commitment of its members (Freelon et al., 2018; Hirsch, 1990). Such an ability can be described as its solidarity. Solidarity has been used to describe both an individual-level and collective-level property. An individual can feel solidarity with a group when they feel committed to it or the cause it supports (Hirsch, 1990). A collective, such as a crowd, can also have high (or low) solidarity, referring to its ability to draw voluntary contributions and concern for its well-being from its members. We use the term in this latter, collective sense. Solidarity in this sense is a latent property of the crowd that is revealed through its aggregate behavior—the actions taken voluntarily by its members and supporters. It is thus an important variable for the crowd’s survival and influence but not easily observed or controlled.

Although incentive and formal control can technically establish solidarity (Hechter, 1988), the two primary informal structures that constitute solidarity are interpersonal bonds and collective identity (Friedkin, 2004; Prentice et al., 1994). Individuals are more willing to volunteer to help collectives that contain many of their personal ties (Friedkin, 2004; McAdam, 1986). They are also more willing to step forward for collectives from which they draw social identity or expect to gain material benefits (Turner, 1975; Wilson and Wilson, 2007).

Both interpersonal bonds and collective identity have been found to be essential to online communities (Ren et al., 2007) and are also critical to the sustenance and effectiveness of social media crowds. These crowds need ongoing participation to survive but lack the ability to incentivize or coerce engagement through formal structures and must compete for attention with many collective entities with which their targeted population of potential participants overlaps (Bimber et al., 2005). Thus, unsurprisingly, engagement with social media crowds can grow quickly but also tends to decay rapidly (Fusaroli et al., 2015; Jungherr and Jurgens, 2014). The question of how some crowds are able to build the informal structures of solidarity that sustain participation, while others fail to do so, is thus critical.

To date, research suggests that participation itself increases solidarity, leading to further participation. There are several mechanisms through which this can occur. In each case, the larger a crowd grows, the more attractive it becomes to both members and bystanders. For example, shared attention, in which individuals are aware that they, and others, are attending to common entities, both draws attention to and encourages sustained engagement with these entities (Shteynberg, 2015). Direct attention from other members of the crowd also sustains crowd members’ engagement (Vaccari et al., 2015). Another mechanism is entitativity, the extent to which a social group is perceived (either by members or outsiders) to be a coherent whole (Campbell, 1958). Entitativity enhances perceptions of a group’s strength and can be enhanced by the extent to which members show similar behavior (Lickel et al., 2000). As we elaborate below, the expression of emotion can also play a similar role in both retaining the engagement of existing members and attracting new members. Thus, we hypothesize the following:

H1. Crowd activity is self-sustaining—the quantity of participation in a crowd at the most recent shared event will predict the quantity of participation at the next shared event.

Crowd solidarity and expressions of discrete emotions

Crowds and emotion

Emotion is an essential feature of crowds. Early, 19th century descriptions of crowds characterize them as “spontaneous social behavior directed by aroused emotion” (Lin, 2015: 236). More recently, research has shown crowds to respond emotionally to events (Lucas et al., 2017), with some arguing that people may seek out social media crowds for the purpose of expressing emotion (Bennett and Segerberg, 2012).

Crowds and emotions are also linked theoretically in that both are described as social responses to surprising or novel information. Specifically, at the individual level, emotions are responses to unexpected information that can initiate social behavior. They provide appraisals of a situation that can quickly guide action, bypassing both slower, more complex deliberation (Tooby and Cosmides, 2008) and the (suddenly mismatched) routines that normally apply (Weick, 1995). Crowds are a common, though by no means exclusive, social phenomenon that results from such a failure of expectations where those who are thrilled or disappointed gather together.

Crowds and emotions are also linked more directly. Many scholars argue that emotions are socially functional (Fischer and Manstead, 2008; Van Kleef, 2016). Emotions permit individuals to convey social information because they enable individuals to communicate and infer one another’s inner states, such as their intentions and construal of social situations (Van Kleef, 2016). They thus stimulate actions and responses in others, enabling rapid social coordination (Turner, 2007). Emotions support cooperation because they are “honest” signals, meaning they are harder to fake than verbal ones (Frank, 1988) that can be easily assessed by others (Reeck et al., 2016). Crowds are social settings in which expressed emotions are highly visible to many people. Even when confined to social media, emotions can be contagious (Coviello et al., 2014; Kramer et al., 2014). Thus, if emotions are socially functional as many have argued, it is likely that they do so with particular importance in crowds, where they are often expressed and easily observed, and where other organizational processes, such as formal control, are limited.

Emotion and solidarity

Unsurprisingly, emotion has been linked directly to solidarity (Knight and Eisenkraft, 2015). For example, members’ affective attachment to a community has been found to be critical to community survival (Kanter, 1968). Particularly, affective convergence or social sharing of emotions, in which group members’ emotional states become increasing similar through interactions (Kelly, 1988; Rimé, 2009), can both engender solidarity and enhance task performance within groups (Collins, 2004; Knight and Eisenkraft, 2015). In settings, where emotions are likely to be volunteered (Bennett and Segerberg, 2012) and contagious (Coviello et al., 2014; Kramer et al., 2014), crowds should benefit from these solidarity-enhancing effects.

Crowds which organize around social identities, or form such identities through the establishment of entitativity, can have solidarity enhanced further. Individuals who identify with a group experience group-based emotions (Mackie et al., 2015). In group-based emotions, individuals respond to events or information that bear on their group as though it were experienced personally, such as a terrorist attack on their nation that occurs thousands of miles away (Lin and Margolin, 2014). Crowds of individuals who share identification with a group or cause can thus expect to be triggered into similar emotions by external events. Upon expressing these emotions, they will reveal their common attitudes to one another, further enhancing solidarity.

The role of discrete emotions

The relationship between crowd solidarity and emotion is further complicated by the discrete nature of some emotions (Nabi, 2010). We focus on four categories of such discrete emotions, which are considered fundamental to more complicated emotions (Turner, 2007). Each orients individuals toward specific behavior responses (Tooby and Cosmides, 2008). Positive emotions are often not distinguished, as their behavioral orientation is similar, generally related to approaching and continuing rewarding states (Fredrickson and Cohn, 2008). By contrast, there are three distinct negative emotions with different behavioral consequences. Anger is a response to an obstacle to goal attainment (e.g. a threat) and often directed to an object (Lemerise and Dodge, 2008), including out-groups (Mackie et al., 2000). Sadness is associated with the appraisal of loss, including loss of attachment and identification to social objects, and often triggers a “slowing down” of cognitive processes associated with closer inspection (Charmaz and Milligan, 2006). Fear–anxiety has the function of driving the individual away from potential hazards (Öhman, 2008).

These different functional roles suggest discrete emotions have different implications for solidarity building. For example, anger is a response to threats associated with certainty and a feeling of group cohesion (Mackie et al., 2015). By contrast, fear–anxiety is associated with a perception of weakness in one’s group or strength in a rival’s (Mackie et al., 2015). Fear–anxiety can also develop into anger as time passes and events become more certain and controlled (Lin and Margolin, 2014), while sadness can give way to anxiety (Doré et al., 2015).

In sum, the various findings described above suggest emotional expression may contribute to the solidarity of crowds. Crowds bring together individuals who are likely to express emotion. This emotional expression can reflect collective identity, and hence its consistency can affect solidarity. The expression of different discrete emotions in close proximity to others can also influence their appraisals of the crowd and related events. Nonetheless, implications of these ideas for specific crowd behavior have not been explored. Below, we articulate a two-step mechanism: shared events trigger emotional expression; emotional expression sustains or suppresses solidarity.

The role of emotional expression: a two-step mechanism

Shared events triggering emotional expression

Social media crowds tend to gather around shared events, such as the formation of real-life crowds (Bastos et al., 2015; Jungherr and Jurgens, 2014), broadcast media events (Lin, Keegan, Margolin, & Lazer, 2014), and breaking news (Toepfl and Piwoni, 2018). These events are “laden with great uncertainty, risk, and drama” (Vaccari et al., 2015: 1042) and often lead to spikes in activity that can lead to sustained engagement (Fusaroli et al., 2015; Jungherr and Jurgens, 2014). These events have also been observed to trigger strong emotions, particularly among those who identify with the crowd as a social group (Cui et al., 2016; Lin and Margolin, 2014). In particular, strong emotions are likely to be stimulated by two aspects of a shared event, that is, valence and unexpectedness.

Valence reflects the change in intergroup status for the crowd as a social group (Turner, 1975). As a result of some events, some groups gain power and status. In some cases, crowds are in explicit competition with one another, such as in sporting events or political contests (Kim et al., 2015; Vaccari et al., 2015). In other cases, competition emerges as a crowd gains attention and power and “counter-movements” rise against it (Freelon et al., 2018; Jensen and Bang, 2013).

The valence of an event is determined by its implication for the crowd’s status in this competition. Events that show that the crowd or its cause is successful or achieving high status should stimulate positive group-based emotions (Cui et al., 2016), while events that show that the crowd or its cause is failing or being relegated to lower status should stimulate negative group-based emotions (Lucas et al., 2017):

H2. Events of group success lead to more expressions of positive emotion (H2a) and fewer expressions of negative emotions (H2b) in crowds, in contrast to events of group failure.

RQ1. How are discrete negative emotions (i.e. anger, sadness, and anxiety–fear) expressed differently in crowds after group failures?

The unexpectedness of an event is also likely to stimulate emotions. Emotions deliver rapid responses that have been inherited as adaptively advantageous (Tooby and Cosmides, 2008). They are thus especially suited to solve short-term coordination problems in acute, unexpected situations when established collective processes (e.g. codified rules and procedures) are less likely to be effective (Weick, 1995). The occurrence of surprising events—those not previously expected—should thus amplify emotional expression:

H3. Unexpected group success leads to more expressions of positive emotions than expected group success (H3a), whereas unexpected group failure leads to more expressions of negative emotions than expected group failure (H3b).

RQ2. How are discrete emotions (i.e. positive emotion, anger, sadness, and anxiety–fear) expressed differently in crowds in relation to the unexpectedness of group outcomes?

Emotion expression and crowd solidarity

Three mechanisms suggest emotional expression can stimulate or suppress solidarity, leading to further participation in the crowd. First, the raw quantity of emotion expressed can indicate the commitment of existing crowd members to the crowd, enhancing entitativity (Mackie et al., 2015). When such emotion is homogeneous, there can also be an enhancement to solidarity. Emotional expression can also spur emotional contagion, increasing homogeneity (Coviello et al., 2014; Kramer et al., 2014).

Second, the qualities of specific, discrete emotions expressed can also influence solidarity. Most simply, positive emotion has been argued to create a self-sustaining “buzz” of positive emotions that psychologically rewards members for continuing participation in a collective activity (Collins, 2004; Kelly, 1988). The discrete negative emotions expressed in the crowd can also influence ongoing participation. Emotional expressions by some crowd members can motivate others to address particular group-oriented concerns, that is, the social appraisal of a group (Fischer and Manstead, 2008). For example, anger can trigger collective action, often toward out-groups (Mackie et al., 2015). Group members’ expressions of fear–anxiety can stimulate collective sense-making, thus inviting more discussion regarding resolutions of such uncertainties (Jensen and Bang, 2013; Weick, 1995).

Finally, crowd members’ emotional expression may elicit interpersonal responses from other group members. For example, when group members express sadness, they can invite sympathetic responses from others (Van Kleef, 2016). In sum, emotional expression is expected to influence further participation, particularly in the short term while emotions are felt intensely, but in a manner that may encourage the development of stronger interpersonal and person–group ties to the crowd. Given the heterogeneity of the four discrete emotions, we also ask the following:

RQ3. How do discrete emotions (i.e. positive emotion, anger, sadness, and anxiety–fear) differently affect the participation in the crowd?

Methods

A natural field experiment

Studying the processes described above requires an operationalization of crowd variables, such as success/failure and expectations of this success that are difficult to measure for many crowds. At the same time, many crowds have different values and goals which may correspond to their tendencies for emotional expression. For example, some crowds may prize criticism and negative affect, whereas others focus more on a constructive, positive approach. To address these challenges, this study uses tweets from NFL fans in response to game results. Sports contests have been used to study various social processes (Frey and Eitzen, 1991), such as group identity expression (Cialdini et al., 1976) and group-based emotion (Lucas et al., 2017). They also offer features particularly useful for this study. First, sports leagues are attended by numerous crowds with comparable goals, enabling the study of a “population” of similar crowds. Sports contests are also frequent and scheduled in advance, allowing the compilation of panel data that is difficult to obtain for other crowds. Second, sports contests offer a clear operationalization of success and failure (winning and losing), as well as a quantitatively precise and comparable set of measurements of expectations for these outcomes (through win probability, described below). Finally, because the game results are largely unknown in advance, they represent quasi-randomized trials of competitive event outcomes. That is, at least within the context of a specific event (game) and a crowd gathered on social media, fans can observe players and game action but players cannot see the social media crowd’s response (until the event is over). Thus, causal effects flow from game results to crowd behavior but not in the reciprocal direction, ensuring that endogenous dynamics do not influence the distribution of treatments (who wins or loses).

Data collection

We collected tweets from the first 11 weeks (10 games played) of NFL 2013–2014 regular season as follows. First, we downloaded the game schedules from pro-football-reference.com . Then, we retrieved in real time 5,438,887 tweets sent during ±1 hours of 130 games played by the 32 teams. In particular, we requested tweets that contained team hashtags or @mentioned the teams’ Twitter handle using Twitter’s Steaming application program interface (API). An error in the code lead to only incomplete data being retrieved for Week 2 and Week 3, and thus these observations were dropped. This also leads to Week 4 being excluded for testing the self-sustenance tendency (H1), as there is no t-1 data (Week 3) to serve as the immediate precedent. After the season, we retrieved lists of followers from the 32 teams’ official accounts using Twitter’s REST (representational state transfer) API. Finally, we downloaded game results, locations, and teams’ pre-game and last-quarter win probabilities for each game as derived from football betting markets from pro-football-reference.com.

Measurements

Valence and unexpectedness of shared events

We treat the binary game result as an event’s valence for a given team (win = 1, loss = 0). The event unexpectedness (mean [M] = .43, standard deviation[SD] = .15) was measured as the valence’s absolute deviation from the pre-game win probability that denotes public expectation (e.g. a team with a .9 win probability would win expectedly, that is, the absolute value of 1 – .9 is .1, but lose unexpectedly, that is, the absolute value of 0 – 0.9 is .9).

Crowd participation quantity and emotional expressions

We operationalized crowd participation as original tweets (i.e. regular tweets and replies) that (a) hashtag or at-mention a team’s name or official account and (b) were sent by exclusive followers—individuals who follow one, and only one, team on Twitter (Margolin et al., 2015). Exclusive followers sent 57.48% of all retrieved tweets, in addition to 27.60% and 14.91% tweets from users following no team and more than one teams, respectively. Among exclusive followers’ tweets, 62.17% were original. We exclude retweets because it is not clear if they represent distinct emotional expressions.

Crowd participation quantity was measured as word count in three 1 hour periods—the pre-game hour, the last hour of a game, and the post-game hour. The medians (N = 260) of the three measures are 4204, 15,006, and 11,092 words (skewness = 2.73/1.03/1.68). To restore normality and simplify analysis, we log-transformed the measures (see Table 1 for descriptive statistics).

Table 1.

Descriptive statistics for tweets averaged across games (N = 260).

	log(Quantity)		Positivity %		Anger %		Sadness %		Anxiety %
	M	SD	M	SD	M	SD	M	SD	M	SD
Original tweets from exclusive followers
Pre-game	8.38	0.93	3.17	0.85	0.50	0.31	0.13	0.12	0.12	0.17
Last-hour	9.53	0.70	3.15	0.57	1.10	0.40	0.27	0.17	0.21	0.12
Post-game	9.19	0.84	3.94	1.01	1.03	0.53	0.33	0.26	0.19	0.13
All tweets from exclusive followers
Pre-game	–	–	2.53	0.74	0.36	0.20	0.10	0.09	0.09	0.11
Last-hour	–	–	2.55	0.55	0.89	0.41	0.21	0.16	0.18	0.10
Post-game	–	–	3.24	1.03	0.84	0.53	0.30	0.28	0.16	0.13
All retrieved tweets
Pre-game	–	–	2.28	0.68	0.32	0.17	0.10	0.09	0.09	0.11
Last-hour	–	–	2.50	0.52	0.79	0.32	0.18	0.11	0.15	0.09
Post-game	–	–	2.92	1.03	0.66	0.38	0.22	0.24	0.13	0.12

M: mean; SD: standard deviation.

Emotional expressions of positive emotion, anger, sadness, and fear–anxiety were measured as percentages of affect words defined by Linguistic Inquiry and Word Count 2015 (LIWC, Pennebaker et al., 2015). To our best knowledge, LIWC is currently the only computerized tool that analyzes discrete emotions, and its psychometric properties have been extensively studied (see, for review, Tausczik and Pennebaker, 2010). Before the measurement, we substituted football-related words that are also LIWC-defined affect words (e.g. plays, win, triumph-, champ-, and strength-) with an undefined character. As mentioned above, emotional expression was only measured using original tweets from exclusive followers. These individuals are most likely to identify with the team and have been shown to tweet accordingly (Margolin et al., 2015) and are less likely to be transient members of the crowd (Jungherr and Jurgens, 2014). This condition enhances our ability to examine hypotheses 2 and 3 and RQs 1 and 2 where emotional expression is the outcome variable, as exclusive followers should be more emotionally responsive to game outcomes than generic accounts. However, this decision can weaken our ability to observe the influence of emotional expression on future participation (RQ3). This is because crowd participants tracking tweets to the team hashtag stream cannot easily distinguish group members from other tweeters, and so may be affected by emotional expression in the stream as a whole. To be sure, our results were robust to these decisions, we also run models for RQ3 on (a) all tweets (including retweets) sent by exclusive followers and (b) all retrieved tweets.

Results

Figure 1 illustrates temporal orders of all measurements. Figure 2 summarizes patterns of crowd participation quantity and emotional expression measures across five periods of an average game, conditioning on game outcome (i.e. win vs loss).

Figure 1.

Study overview. Vertical lines and gray area indicate measurement times and periods.

Figure 2.

Temporal changes of participation quantity and emotional expressions (original tweets from exclusive followers). Each thin line is one team averaged across games. Thick lines are polynomial fitted lines (cubic) across 32 teams. Statistics were summarized within 1 hour before the game, three-thirds (each about 1 hour roughly) during the game, and 1 hour after the game.

Self-sustenance of participation in crowds

To test H1, we regressed the pre-game participation quantity at game week t(Q^(pre)) on the post-game participation quantity at game week t-1 (Q^(post)). We used a two-way (team × week) fixed-effects model to control for team characteristics and general characteristics of each game week for all teams, along with potentially confounding covariates. Team characteristics can lead to fan self-selection because fans of a strong team may be more conversational and affectively engaged in all games. As a result, crowd participation across games can be spuriously correlated. We included the week t game’s pre-game win probability (W, centered at .5) to control for fan expectations because previous victories may influence both post-game participation and make fans more optimistic about the next game. We included a binary indicator of game time (P, prime time = 1) in relation to prime time (8–10 p.m.) because prime time games likely receive larger audiences and hence affect pre-game participation quantity. We also included a binary indicator of game location (H, home = 1) as games played at home field may attract more attention from home-team fans. The regression is summarized below (N = 196, n_i = 32, ${\bar{n}}_{t} = 6.1$ , standard errors [SEs] in parentheses are robust to clustering):

\begin{array}{l} Q_{i t}^{(pre)} = 0.09 Q_{i, t - 1}^{(post)} + 3.53 W_{i t} - 0.41 W_{i t} Q_{i, t - 1}^{(post)} + 0.80 P_{i t} + 0.17 H_{i t} + α_{i} + β_{t} + e_{i t} \\ (0.04) (1.87) (0.20) (0.05) (0.06) \end{array}

The post-game quantity indeed predicts the following pre-game quantity (t[31] = 2.18, p = .037, η² = .025), thus supporting H1. The interaction between win probability and post-game quantity (Q^(pre)W) is included and suggest the self-sustaining tendency is stronger (b = 0.16, standard error (SE) = 0.06, p = .007, η² = .04) when a team is expected to win (i.e. win probability at +1SD).

Effects of shared events on emotional expressions

We regressed post-game expression of each of the discrete emotions on an event’s valence, unexpectedness, and their interaction. We used a similar two-way fixed-effects model as described above because team characteristics would induce spurious correlation between game result and emotional expression because fans of a team that generally wins a lot of games may be more affectively engaged than fans of other teams. We then controlled for a team’s pre-game win probability, as it can similarly induce selection, with fans more likely to attend to games when their team is expected to win. We also controlled for the game location. This is because, although win-probability controls for the effect of the “home-field advantage,” the emotional profiles or temporal patterns of home-team fans’ participation may differ due to the greater likelihood of them being physically present near the game. Finally, we controlled for the measure of emotional expression before a game starts as a baseline. The models are summarized in Table 2.

Table 2.

Summaries of fixed-effect regressions predicting emotional expressions (post-game).

	Positive			Angry			Sad			Anxious
	Raw	Std.Y	η²	Raw	Std.Y	η²	Raw	Std.Y	η²	Raw	Std.Y	η²
Valence (win = 1)	1.10*** (0.16)	1.08*** (0.15)	.32	−0.75*** (0.08)	−1.43*** (0.16)	.49	−0.42*** (0.03)	−1.65*** (0.11)	.62	−0.15*** (0.02)	−1.13*** (0.13)	.29
(SEM estimation)	1.10*** (0.15)	1.08*** (0.15)		$- {0.75}_{(a)}^{* * *} (0.08)$	$- {1.43}_{(b)}^{* * *} (0.15)$		$- {0.42}_{(a)}^{* * *} (0.03)$	$- {1.65}_{(b)}^{* * *} (0.10)$		$- {0.15}_{(a)}^{* * *} (0.02)$	$- {1.14}_{(b)}^{* * *} (0.12)$
Unexpectedness × Valence	1.78* (0.80)	1.75* (0.78)	.02	−0.22 (0.46)	−0.42 (0.87)	<.01	−0.32 (0.17)	−1.26 (0.66)	.02	−0.31** (0.10)	−2.33** (0.75)	.03
(SEM estimation)	$- {1.76}_{(c)}^{*} (0.77)$	$- {1.73}_{(d)}^{*} (0.76)$		−0.24 (0.45)	−0.46 (0.85)		−0.32(c) (0.17)	−1.27(d) (0.65)		$- {0.31}_{(c)}^{* *} (0.10)$	$- {2.38}_{(d)}^{* *} (0.74)$
Unexpectedness (centered at .5)	−0.65 (0.48)	−0.64 (0.47)	<.01	0.63* (0.30)	1.19* (0.57)	.04	0.28* (0.12)	1.10* (0.47)	.01	0.16* (0.07)	1.20* (0.50)	<.01
Relative location (home = 1)	0.08 (0.09)	0.08 (0.09)	<.01	0.03 (0.05)	0.06 (0.10)	<.01	0.01 (0.02)	0.06 (0.09)	<.01	−0.01 (0.01)	−0.07 (0.10)	<.01
DV baseline (pre-game)	0.10 (0.06)	0.10 (0.06)	.01	0.01 (0.10)	0.01 (0.18)	<.01	−0.09 (0.07)	−0.34 (0.28)	<.01	0.04 (0.04)	0.30 (0.27)	<.01
R²_{all/between/within}	.34/.27/.37			.50/.37/.53			.66/.71/.65			.33/.43/.31

SEM: structural equation model; DV: digital video; Std.Y: standardized response; SE: standard errors; df: degrees of freedom.

N = 260 (two-way panels n_team = 32, ${\bar{n}}_{week} = 8.1$ ). SEs (adjusted for clustering in teams) are parenthesized. The within/between R²s reflect the team dimension. Only estimates (est.) of interest from the SEM are displayed. Subscripts (a) to (d) indicate equality constraints on the SEM with $χ_{(a)}^{2} (2) = 159.25$ , $χ_{(b)}^{2} (2) = 20.48$ , $χ_{(c)}^{2} (2) = 3.73$ , and $χ_{(d)}^{2} (2) = 2.94$ .

p < .05; **p < .01; ***p < .001 (based on t-distribution with df = 31).

Per H2, a success should lead to more positive (H2a) yet less negative emotional expressions (H2b). Estimates in Table 2 support the hypothesis at α_{Βονφερρονι} ≤ .05 (i.e. p ≤ .0125 for four tests): At a neutral unexpectedness, a success significantly leads to more positive expressions (p < .001, η² = .32) but less negative expressions of anger, sadness, and anxiety (ps < .001, η²s = .49/.62/.29).

RQ1 further asks if the preceding effects differ across angry, sad, and anxious expressions. We estimated the same regressions of the three negative expressions, along with that of positive expression for completeness, as a structural equation model (SEM; χ²[12] = 20.53, p = .058; comparative fit index [CFI] = .99; root mean square error approximation [RMSEA] = .05 with 95% confidence intervals [CIs] = [0.00, 0.09]) with the two-way fixed-effects entered as dummy variables. To answer RQ1, we tested parameter constraints on the SEM in a standardized metric (i.e. standardized endogenous variables; Table 2 also lists estimates in the raw metric). This is because we care about practical differences in an event’s effect on the four emotional expressions, which have very different SDs (0.13–1.01). The standardized effects of event valence on angry, sad, and anxious expressions are −1.43, −1.65, and −1.14. Equally constraining the three effects in the SEM worsens its fit by χ²(2) = 20.48, p < .001, suggesting a significant overall difference. Pair-wise tests reveal the significant difference is likely due to that between the effects on sadness and anxiety (p < .001) but not the others (p_{anger-sadness} = .090, p_{anger-anxiety} = .047) after correcting a family-wise error rate at α_Bonferroni ≤ .05 (i.e. p ≤ .0167).

H3 postulates the unexpectedness of an event should moderate effects of event valence on emotional expressions. H3a predicts this effect to be positive on positive expression and H3b predicts it will be negative on negative expressions. Consistent with the prediction, the interaction effect is significantly negative on expression of anxiety (p = .004, η² = .03) and marginally for sadness (p = .066, η² = .02), and significantly positive on expression of positive emotion (p = .033, η² = .02). In other words, unexpected losses lead to the expression of additional anxiety and likely sadness as well, whereas unexpected wins lead to the expression of more positive emotion. The effects are non-significant on anger (p = .636, η² = .002). We thus consider H3 as partially supported.

RQ2 asks if the moderating effect of event unexpectedness differs across the four emotional expressions. Equally constraining the absolute value of the three significant effects (positive, anxiety, and sadness) in either the standardized metric or the raw metric does not significantly worsen the SEM’s model fit (χ²(2) = 2.94/3.73, p = .230/.155), suggesting little evidence of a difference in the strength of moderation for the three emotions. Anger is different, as it is non-significant as described above with a very small effect size (η² = .002, 90% CI = [0.000, 0.022]).

Effects of emotional expressions

RQ3 concern whether and how the last-hour emotional expressions influence post-game crowd participation, which, as H1 points out, sustains future participation as a positive feedback loop. The similar fixed-effects model is used to control team characteristics, as described before. To further reduce spurious correlations between the last-hour emotional expressions and the post-game participation quantity, we control for as many characteristics of a particular game as possible. These include (a) event valence, event unexpectedness, and their interaction, (b) the win probability at the fourth quarter of a game, and (c) the game location. We also control for a baseline of participation quantity measured for the same period of time as the emotional expressions.

As summarized in Table 3, emotional expression indeed predicted crowd participation, with most effects robust to the inclusion of retweets and tweets from non-exclusive followers. The effect sizes of negative emotional expression (η²s ranged from .03 to .09) are even greater than those of event valence (η²s ranged from .00 to .02).

Table 3.

Emotion expressions predicting post-game participation quantity.

	Original tweets from exclusive followers				All tweets from exclusive followers				All retrieved tweets
	Margin	Loss	Win	η²	Margin	Loss	Win	η²	Margin	Loss	Win	η²
Positive emotion	−0.07^* (0.03)	0.01 (0.04)	−0.16^*** (0.04)	.09	−0.02 (0.03)	0.00 (0.04)	−0.04 (0.04)	.01	0.04 (0.03)	0.08^* (0.04)	−0.01 (0.04)	.02
Angry	0.11^** (0.03)	0.10^* (0.04)	0.12 (0.06)	.04	0.15^*** (0.04)	0.15^** (0.05)	0.15^** (0.05)	.06	0.16^*** (0.04)	0.13^* (0.05)	0.19^** (0.05)	.08
Sad	−0.12^** (0.04)	−0.15^*** (0.03)	−0.08 (0.06)	.07	−0.12^* (0.05)	−0.11^* (0.05)	−0.14 (0.09)	.03	−0.17^*** (0.04)	−0.15^** (0.05)	−0.18^* (0.08)	.07
Anxious	0.00 (0.03)	−0.10^* (0.04)	0.11^* (0.04)	.06	−0.02 (0.03)	−0.17^*** (0.05)	0.14^** (0.05)	.09	−0.03 (0.03)	−0.14^*** (0.03)	0.08^* (0.03)	.06
Unexpectedness (centered at .5)	0.89^*** (0.19)	0.51^* (0.22)	1.27^*** (0.33)	.12	0.78^*** (0.18)	0.55^* (0.22)	1.01^** (0.30)	.09	0.82^*** (0.18)	0.62^* (0.25)	1.01^** (0.29)	.10
Valence (win = 1)	0.30^** (0.09)	–	–	.02	0.29^* (0.12)	–	–	<.01	0.24^* (0.09)	–	–	<.01
Game location	0.02 (0.05)	–	–	<.01	0.02 (0.05)	–	–	<.01	0.03 (0.05)	–	–	<.01
(home = 1)	0.02 (0.05)				0.02 (0.05)				0.03 (0.05)
log(Quantity_last-hour)	0.85^*** (0.08)	–	–	.36	0.84^*** (0.09)	–	–	.35	0.77^*** (0.09)	–	–	.29
Win probability_4Q	0.20^* (0.09)	–	–	.02	0.09 (0.10)	–	–	<.01	0.06 (0.10)	–	–	<.01
R ² _{all/between/within}	.78/.86/.65				.76/.86/.62				.77/.86/.63

SE: standard errors.

N = 260 (n_team = 32, ${\bar{n}}_{game} = 8.1$ ). SEs in parentheses are adjusted for clustering in teams. Emotional expressions were standardized. Interactions with event valence were parameterized as separate slopes, and marginal effects were also estimated.

p < .05; **p < .01; ***p < .001.

Particularly, we found, first, expression of anger stimulates participation regardless of event valence (bs ranged from 0.10 to 0.19). Second, sad expression suppresses participation, especially after a loss (bs ranged from −0.11 to −0.15). Finally, we find a “complementary” pattern in the effects for expression of positive emotion and fear–anxiety. Positive expression appeared to stimulate participation after a loss (bs ranged from 0.00 to 0.08) but suppress participation after a win (bs ranged from −0.01 to −0.16). In a similar but reversed pattern, fear–anxiety expression stimulates participation in the crowd after a win (bs ranged from 0.08 to 0.11) but suppresses it after a loss (bs ranged from −0.10 to 0.14). We note that the effects of positive expression are less consistent across the three measurement sets, compared to those of negative expressions.

Discussion

This study explores the relationship between emotional expression and solidarity in social media crowds and finds evidence that crowds may gain or lose solidarity through the emotions expressed by their members in response to shared events. These findings have implications both for the study of social media crowds as well as the study of emotional expression.

The study finds preliminary support for a generic mechanism of emergent self-organization that links the micro- and macro-level realities of social media crowds. Specifically, crowd members’ shared experience at the macro-level (a shared event) spurs individual behaviors at the micro-level (the expression of emotions) which in turn stimulate or suppress a macro-level property of the crowd (solidarity). This mechanism is generic in that these dynamics are based on general properties of crowds (shared attention and mutual exposure) and psychological tendencies of individuals (emotional expression) and is not necessarily guided by mindful reasoning by crowd members explicitly seeking to build solidarity. In fact, crowd members may engage with the event for personal purposes—such as for entertainment—and experience their emotions as authentic, individual-level experiences, without awareness that they serve this collective, self-organizing function.

The question of how, exactly, emotional expression builds solidarity remains open, however, and its investigation will require more detailed study at both the macro- and micro-level. Prior research suggests several related avenues: the experience of shared emotional expression enhances an individual’s identification with the crowd as in an “interaction ritual chain” (Collins, 2004) or builds interpersonal bonds with group members through affective “entrainment” (Kelly, 1988). Alternatively, the distribution of emotional expressions within a crowd may affect perceived similarity and common fate, that is, the entitativity of the crowd, in the eyes of perceivers (Campbell, 1958), making the crowd attractive or unattractive for further social engagement. Future work should attempt to tease part these differences, in particular, by focusing on the ongoing participation decisions of specific individuals in relation to their exposure to the crowd’s emotions through both experimental and observational data.

Research might also investigate how crowds come to express specific discrete emotions, as our findings indicate that differences in expression influence the crowd’s development. In particular, though emotions are felt and expressed by individuals, to what extent do individuals consider the organizational properties of the crowd when responding emotionally to shared events? For example, prior research suggests that anger is often expressed when there is a sense of an addressable threat, while sadness is engendered as a reaction to helplessness. Perhaps, these appraisals are made with reference to features of the crowd itself. For example, perhaps anger expression is triggered when crowd members’ perceive untapped strength in the crowd, spurring it to address a threat, whereas sadness is triggered by a sense that the crowd is in need of repair. Alternatively, perhaps anger and sadness are contagious to differing degrees, with one more reliant on being expressed within key roles or positions within the crowd to become dominant.

Finally, though they should first be replicated on social movement crowds specifically, our findings suggest that the study of solidarity in these specific crowds should consider the distinct role of emotional expression. In particular, the social movement crowd literature often focuses on solidarity through unity or consensus, such as the sharing of narratives or frames (Bennett et al., 2011; Freelon et al., 2018). Our results suggest that some forms of solidarity may be sparked before this more sophisticated unity is formed. In particular, our results show that raw emotional expression immediately following an event, before consensus is likely to have emerged, has an impact on solidarity. This suggests narratives and frames, while possibly critical in the long run, may be products of other, more basic solidarity building processes.

Limitations

These arguments also point to some limitations of the study. Most prominently, we measure all variables at the crowd level. This enables us to characterize phenomena in terms of collective antecedents and consequences; however, we lack the granularity to distinguish micro-level mechanisms that might give rise to these macro-phenomena. For example, crowd-level emotional expressions may be the aggregation of a common manifestation of individual emotional expressions, that is, when the group fails, most or many members feel and express sadness. It is also possible, however, that only a small number of individuals are initially triggered in this way, and that as leaders or through emotional contagion their expressions spread to others. Further research should attempt to arbitrate between these explanations.

Another limitation is our binary operationalization of group identifiers as “exclusive followers” without regard for the intensity of that identification. Such intensity may be theoretically important, particularly in the context of fluid group membership. For example, strong identifiers may feel there is no option to abandoning group identity after group failures, and so may be more likely to express anger to address or root out threats, whereas weak identifiers may exit or, perhaps, experience sadness as a response that encourages them to reconsider their membership. Future research should thus consider a more refined measurement of identity strength.

A third limitation is related to the measures of emotional expression in tweets. As described in Table 1, the proportions of affect words defined by LIWC were low in the analyzed texts, suggesting potential reliability issues induced by noise and the sparsity of signals of emotional expression. LIWC has also been criticized for not considering lexical items such as acronyms, initialisms, emoticons, or slang that are prevalently used to express emotion in social media (Hutto and Gilbert, 2014). Nevertheless, the 2015 version of LIWC used in this study has included acronyms and puncture-based emoticons that are frequently used to express emotions in social media and text messaging (Pennebaker et al., 2015). Future research may incorporate more advanced methods, for example, machine learning and human coding, to better capture the sentiments in social media texts.

Finally, we recognize that professional football fandom is itself part of an institutional field that likely contains its own logics (Stohl, 2014). It is thus possible that the effects we observe are not part of a general crowd tendency but one that follows from a unique culture shared by NFL fans. Further research should consider replicating these results in other fields, both within and outside of professional sports, including political campaigns and social movements in which competition is prominent.

Conclusion

This study addresses the role of emotional expression in sustaining participation in crowds. Our findings demonstrate a unique role of emotional expression in influencing crowd solidarity, and show that this influence depends on the discrete emotions expressed. We believe accounting for emotional expression can facilitate the understanding of collective action through crowds, where communication is increasingly crucial to organizing in contemporary societies.

Footnotes

Acknowledgements

The authors would like to thank Yu-Ru Lin for her help with collecting the data and the anonymous reviewers for their helpful feedback.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work is part of research supported by NSF Grants #1634702 and #1563705.

ORCID iD

Drew Margolin

Wang Liao

Author biographies

Drew Margolin is an assistant professor in the Department of Communication at Cornell University. His research focuses on collective processes, particularly related to the production of discourse, that can be analyzed using digital, observational data.

Wang Liao is a PhD candidate in Communication at Cornell University. His work explores the social functionality of communication in interpersonal and group processes, with a particular interest in communication dynamics and structures, affective processes, and technological modulation of communication.

References

Bastos

Mercea

Charpentier

(2015) Tents, tweets, and events: the interplay between ongoing protests and social Media. Journal of Communication 65(2): 320–350.

Bennett

Segerberg

(2012) The logic of connective action: digital media and the personalization of contentious politics. Information, Communication & Society 15(5): 739–768.

Bennett

Foot

Xenos

(2011) Narratives and network organization: A comparison of fair trade systems in two nations. Journal of Communication 61(2): 219–245.

Bennett

Segerberg

(2011) Digital media and the personalization of collective action. Information, Communication & Society 14(6): 770–799.

Bimber

Flanagin

Stohl

(2005) Reconceptualizing collective action in the contemporary media environment. Communication Theory 15(4): 365–388.

Campbell

(1958) Common fate, similarity, and other indices of the status of aggregates of persons as social entities. Behavioral Science 3(1): 14–25.

Charmaz

Milligan

(2006) Grief. In: Stets

Turner

(eds) Handbook of the Sociology of Emotions. New York: Springer, pp. 516–543.

Cialdini

Borden

Thorne

et al . (1976) Basking in reflected glory: three (football) field studies. Journal of Personality and Social Psychology 34(3): 366–375.

Collins

(2004) Interaction Ritual Chains. Princeton, NJ: Princeton University Press.

10.

Coviello

Sohn

Kramer

ADI

et al . (2014) Detecting emotional contagion in massive social networks. PLoS ONE 9(3): e90315.

11.

Cui

Rui

(2016) From immediate community to imagined community: social identity and the co-viewing of media event. Global Media and China 1(4): 481–489.

12.

Doré

Ort

Braverman

et al . (2015) Sadness shifts to anxiety over time and distance from the national tragedy in Newtown, Connecticut. Psychological Science 26(4): 363–373.

13.

Fischer

Manstead

(2008) Social functions of emotion. In: Lewis

Haviland-Jones

Barrett

(eds) Handbook of Emotions. New York: Guilford Press, pp. 456–468.

14.

Frank

(1988) Passions within Reason: The Strategic Role of the Emotions. New York: W. W. Norton & Company.

15.

Freelon

McIlwain

Clark

(2018) Quantifying the power and consequences of social media protest. New Media & Society 20(3): 990–1011.

16.

Fredrickson

Cohn

(2008) Positive emotions. In: Lewis

Haviland-Jones

Barrett

(eds) Handbook of Emotions. New York: Guilford Press, pp. 777–797.

17.

Frey

Eitzen

(1991) Sport and society. Annual Review of Sociology 17(1): 503–522.

18.

Friedkin

(2004) Social cohesion. Annual Review of Sociology 30: 409–425.

19.

Fusaroli

Perlman

Mislove

et al . (2015) Timescales of massive human entrainment. PLoS ONE 10(4): e0122742.

20.

Hechter

(1988) Principles of Group Solidarity. Berkeley, CA; Los Angeles, CA: University of California Press.

21.

Hirsch

(1990) Sacrifice for the cause: Group processes, recruitment, and commitment in a student social movement. American Sociological Review 55(2): 243–254.

22.

Hutto

Gilbert

(2014) VADER: a parsimonious rule-based model for sentiment analysis of social media text. In: Proceedings of the eighth international AAAI conference on weblogs and social media. Available at: https://www.aaai.org/ocs/index.php/ICWSM/ICWSM14/paper/view/8109 (accessed 28 November 2017).

23.

Jackson

Foucault Welles

(2015) Hijacking #myNYPD: social media dissent and networked counterpublics. Journal of Communication 65(6): 932–952.

24.

Jensen

Bang

(2013) Occupy Wall Street: a new political Form of movement and community? Journal of Information Technology & Politics 10(4): 444–461.

25.

Jungherr

Jurgens

(2014) Through a glass, darkly: tactical support and symbolic association in Twitter messages commenting on Stuttgart 21. Social Science Computer Review 32(1): 74–89.

26.

Kanter

(1968) Commitment and social organization: a study of commitment mechanisms in utopian communities. American Sociological Review 33(4): 499–517.

27.

Kelly

(1988) Entrainment in individual and group behavior. In: McGrath

(ed.) The Social Psychology of Time: New Perspectives, vol. 91. Sage Focus ed. Thousand Oaks, CA: SAGE, pp. 89–110.

28.

Kim

Keegan

et al . (2015) Social media dynamics of global co-presence during the 2014 FIFA World Cup. In: Proceedings of the 33rd annual ACM conference on human Factors in computing system, Seoul, South Korea, 18–23 April, pp. 2623–2632. New York: ACM.

29.

Knight

Eisenkraft

(2015) Positive is usually good, negative is not always bad: the effects of group affect on social integration and task performance. Journal of Applied Psychology 100(4): 1214–1227.

30.

Kramer

ADI

Guillory

Hancock

(2014) Experimental evidence of massive-scale emotional contagion through social networks. Proceedings of the National Academy of Sciences of the United States of America 111(24): 8788–8790.

31.

Lemerise

Dodge

(2008) The development of anger and hostile interactions. In: Lewis

Haviland-Jones

Barrett

(eds) Handbook of Emotions. New York: Guilford Press, pp. 730–741.

32.

Lickel

Hamilton

Wieczorkowska

et al . (2000) Varieties of groups and the perception of group entitativity. Journal of Personality and Social Psychology 78(2): 223–246.

33.

Lin

Y-R

(2015) Event-related crowd activities on social media. In: Gonçalves

Perra

(eds) Social Phenomena: From Data Analysis to Models. Cham: Springer, pp. 235–250.

34.

Lin

Y.-R.

Keegan

Margolin

Lazer

(2014). Rising Tides or Rising Stars?: Dynamics of Shared Attention on Twitter during Media Events. PLoS ONE, 9(5), e94093. https://doi.org/10.1371/journal.pone.0094093

35.

Lin

Y-R

Margolin

(2014) The ripple of fear, sympathy and solidarity during the Boston bombings. EPJ Data Science 3(1): 31.

36.

Lucas

Gratch

Malandrakis

et al . (2017) GOAALLL! Using sentiment in the world cup to explore theories of emotion. Image and Vision Computing 65: 58–65.

37.

McAdam

(1986) Recruitment to high-risk activism: the case of Freedom Summer. American Journal of Sociology 92(1): 64–90.

38.

Mackie

Devos

Smith

(2000) Intergroup emotions: explaining offensive action tendencies in an intergroup context. Journal of Personality and Social Psychology 79(4): 602–616.

39.

Mackie

Maitner

Smith

(2015) Emotion and intergroup relations. In: Kosslyn

(ed.) Emerging Trends in the Social and Behavioral Sciences. Hoboken, NJ: John Wiley & Sons. DOI: 10.1002/9781118900772.etrds0111

40.

Margolin

Liao

Lin

Y-R

(2015) Conversing in reflective glory: A systematic study using National Football League games. In: Proceedings of the Ninth International AAAI Conference on Web and Social Media, Oxford, UK: University of Oxford, pp. 646–649.

41.

Monge

Contractor

(2003) Theories of Communication Networks. New York: Oxford University Press.

42.

Nabi

(2010) The case for emphasizing discrete emotions in communication research. Communication Monographs 77(2): 153–159.

43.

Öhman

(2008) Fear and anxiety. In: Lewis

Haviland-Jones

Barrett

(eds) Handbook of Emotions. New York: Guilford Press, pp. 709–729.

44.

Papacharissi

De Fatima Oliveira

(2012) Affective news and networked publics: the rhythms of news storytelling on #Egypt. Journal of Communication 62(2): 266–282.

45.

Pennebaker

Booth

Boyd

et al . (2015) Linguistic Inquiry and Word Count: LIWC2015. Austin, TX: Pennebaker Conglomerates. Available at: www.LIWC.net

46.

Prentice

Miller

Lightdale

(1994) Asymmetries in attachments to groups and to their members: distinguishing between common-identity and common-bond groups. Personality and Social Psychology Bulletin 20(5): 484–493.

47.

Reeck

Ames

Ochsner

(2016) The social regulation of emotion: an integrative, cross-disciplinary model. Trends in Cognitive Sciences 20(1): 47–63.

48.

Ren

Kraut

Kiesler

(2007) Applying common identity and bond theory to design of online communities. Organization Studies 28(3): 377–408.

49.

Rimé

(2009) Emotion elicits the social sharing of emotion: theory and empirical review. Emotion Review 1(1): 60–85.

50.

Shteynberg

(2015) Shared attention. Perspectives on Psychological Science 10(5): 579–590.

51.

Stohl

(2014) Crowds, clouds, and community. Journal of Communication 64(1): 1–19.

52.

Sutton

Gibson

Phillips

et al . (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.

53.

Tausczik

Pennebaker

(2010) The psychological meaning of words: LIWC and computerized text analysis methods. Journal of Language and Social Psychology 29(1): 24–54.

54.

Toepfl

Piwoni

(2018) Targeting dominant publics: how counterpublic commenters align their efforts with mainstream news. New Media & Society 20(5): 2011–2027.

55.

Tooby

Cosmides

(2008) The evolutionary psychology of the emotions and their relationship to internal regulatory variables. In: Lewis

Haviland-Jones

Barrett

(eds) Handbook of Emotions. 3rd ed. New York: Guilford Press, pp. 114–137.

56.

Turner

(1975) Social comparison and social identity: some prospects for intergroup behaviour. European Journal of Social Psychology 5(1): 1–34.

57.

Turner

(2007) Human Emotions: A Sociological Theory. New York: Routledge.

58.

Vaccari

Chadwick

O’Loughlin

(2015) Dual screening the political: media events, social media, and citizen engagement. Journal of Communication 65(6): 1041–1061.

59.

Van Kleef

(2016) The Interpersonal Dynamics of Emotion: Toward an Integrative Theory of Emotions as Social Information. Cambridge: Cambridge University Press.

60.

Weick

(1995) Sensemaking in Organizations. Thousand Oaks, CA: SAGE.

61.

Wilson

(2007) Rethinking the theoretical foundation of sociobiology. The Quarterly Review of Biology 82(4): 327–348.