Psychology has examined emotions primarily as an individual-level phenomenon. Even when considering group-based emotions, an individual's emotional experiences are regarded as an outcome of their psychological belonging or social membership to the given group (Mackie et al., 2000; Mackie and Smith, 2018). Before the individualization of social psychology (Farr, 1980), there had been considerable interest in collective-level psychological phenomena, including emotions, their transference, and their impact on individuals. For instance, Le Bon (1895, p. na) noted that crowds develop ideas and sentiments, and a “collective mind is formed, doubtless transitory, but presenting very clearly defined characteristics”. Similarly, Durkheim (1915) discussed emotional effervescence and collective emotional excitement in the totemic ceremonies of the Uluru and Kingilli people. More recently, there has been a renewed interest in how group memberships shape individual experiences of emotions due to social identification or bonds between members of the group (Kim, 2016). Despite their name, these emotions remain a type of individual emotion - the individual is still the unit of assessment as the appraisal of the event still remains tied to the individual (Goldenberg et al., 2020).

On the other hand, in situations such as wars or national tragedies, an examination of macrolevel affective processes beyond the smaller group-based emotions becomes necessary (Goldenberg et al., 2020). This has led to the development of the notion of collective emotions (CEs), which do not associate the experience of emotions with group memberships. Instead, they consider that people belonging to a collective may experience an emotion for a range of reasons, not limited to shared identities or bonds. CEs arise in a society in response to events such as wars, national tragedies and triumphs, or indeed other events that dilute the boundaries of identification between groups (Bar-Tal et al., 2007; Goldenberg et al., 2014). Research has shown that CEs tend to cut across identities to engulf entire collectives. For instance, Garcia and Rimé (2019) examined Twitter posts of over 60,000 users in the aftermath of the terrorist attacks in Paris in 2015 and found a long-term increase in lexical indicators for solidarity and pro-sociality. CEs emerge not only in real time but also when collectives recall a significant event. For instance, in Kim's (2016) study several years after 9/11, CEs developed amongst participants discussing the trauma of the terrorist attacks.

In other words, CEs emerge in broader group contexts and, unlike individual level emotions, reflect the emotional state of a collective responding to the same event, phenomenon, or situation. CEs capture the overall emotional states of the systems where individuals activate, amplify, or reflect each other's emotional experiences, resulting in a predominant emotional state. Let us take a hypothetical event that has the possibility of evoking emotional experience in a collective: India winning the Football World Cup. If the collective at hand is of n individuals who watched the final separately on their TVs and remained in a communicative black hole, each of them would individually appraise the event and have an emotional experience. However, in a world where people interact, the emotional experience of the event will also be driven by the experiences of others. In a cricket-mad country, Person A may not care about what the country achieves in football. Yet, by virtue of being embedded in the social world where others do, they may eventually experience joy, pride, or other similar emotions. In summary, when a collective system develops a certain emotional state, individuals activate each other, and instead of individual appraisals, the collective state becomes the relevant trigger for emotional experiences. Figure 1 demonstrates these points. What is more, the Internet and modern technology, more than ever before in human history, have enabled near-real-time conversations about events in the social world. This makes the emerging field of CEs extremely pertinent to what Sheldon Stryker referred to as sociological social psychology. This study makes two significant contributions in this field. First, it makes a contribution by outlining an approach to measuring CEs from what people post on the Internet, and developing daily estimates of CEs in large collectives. Its second and more important contribution is of developing a novel application of CEs to explore the embeddedness of individuals in the collective and the social. The research develops a new approach to identifying events in the social world that have significant impacts on the emotional states of collectives.

Three main methodological issues needed to be tackled in the project. First, an event sufficiently powerful to lead to the development of collective emotions needed to be identified. The COVID-19 pandemic adequately fits this criterion. The spread of the disease had an unprecedented impact on life globally and was marked by almost daily changes to public advisories in most countries. Lockdowns and restrictions on civil liberties affected the work and financial circumstances of people across the world (Bell and Blanchflower, 2020; Brewer and Gardiner, 2020; Nicola et al., 2020) and altered social norms and everyday lives (Alon et al., 2020; Dryhurst et al., 2020; Haleem et al., 2020; Nivette et al., 2020; Tagat and Kapoor, 2020; van der Westhuizen et al., 2020). More than any other new infectious diseases of the recent past, the rapid spread of COVID-19 was a cause of tremendous anxiety, alarm, and uncertainty (Goodwin et al., 2011; Mayor et al., 2013; Idoiaga Mondragon et al., 2017) around the world, and a number of studies have noted the pandemic to be an extremely emotional event (Andrade, 2020; Galea et al., 2020; Groarke et al., 2020). Given the high levels of public uncertainty and near daily changes in government advisories early in the pandemic, the study looked at CEs during the first 120 days¹ of the COVID-19 pandemic.

Second, a source of data for estimating collective emotions in response to the said event needed to be identified as CEs develop out of interactions and communications between people dealing with a phenomenon of collective significance. Most debates, discussions, and conversations about the pandemic took place on the Internet as social restrictions and the nature of the disease prevented people meeting others in real life. One of these online public spaces was provided by Twitter, which is widely used for exchanging news, opinions, and feelings in real time. An online space such as Twitter, which people can access at all times, provides a natural means for the contagion, amplification, and reactivation processes and properties of CEs (Goldenberg et al., 2014, 2020). To illustrate, a person expresses their anxiety about the pandemic at time t. Their anxiety decays naturally, like all individual-level emotions, but on reengaging with Twitter, where the CE of anxiety remains high, they may experience a reactivation of their own anxiety. These reasons made Twitter the ideal landscape for examining the development of CEs, and not surprisingly, it has been used in other research on CEs previously (e.g., Garcia and Rimé, 2019).

Third, the project uses countries as the level of the collective at which CEs were tracked. The spread of the disease, hospitalizations, infections, and mortality rates were most commonly discussed at the level of countries. Similarly, the response to the pandemic—lockdowns, restrictions on civil liberties, support for workers and families—also took shape at the level of countries. For these reasons, this research considered countries to be the most appropriate level of the collective for studying CEs in response to the pandemic.

The following sections describe the methodological steps and analytic procedures.

The work presented in this study is among the first to combine big-data research, natural language processing techniques, and time series analysis to examine collective psychological phenomena. It is also among the first to examine collective emotions in the context of a global phenomenon and study them during the same time period in multiple countries. To the best of the authors' knowledge, this research is the first reported work to move away from survey methodologies to identify significant emotional events within and across countries by using naturally occurring text on Twitter. These methodological developments and the practical application of collective emotions were one of the main motivations for sharing this work with our peers.

Due to resource constraints, the current study did not examine the text of the Twitter posts beyond using LIWC to identify lexical indicators for anxiety and positive emotions. Future studies could, for example, also use topic modeling to scaffold the identification of the main topics as a way of validating the events that explain the shock. Due to the large number of data points and individual posters in the dataset, we did not filter for bots. However, the texts posted by bots and messages with false rumors and misinformation are no different in their contributions to the development of collective emotions.

Findings of this research have implications for public policy during periods of disasters and national crisis. As reported, in five countries, the announcement of economic support packages by their respective governments helped reduce anxiety and enhance positive emotions amidst the pandemic. It is likely that the announcement of economic support packages allowed people to come to terms with job losses and other financial challenges posed by the pandemic. This finding emphasizes the significance of providing economic security to the public during times of unprecedented uncertainty and disruption. A very important finding from this project is regarding the impact of the WHO's declaration of a pandemic on public emotions. The WHO was heavily criticized for declaring H1N1 swine flu as a pandemic in 2009 (Godlee, 2010; Low and McGeer, 2010), but with COVID-19, its reluctance in using the term received significant attention (Green, 2020; Vogel, 2020). In part, the reluctance reflected the conventional wisdom that terms such as “pandemic” and “epidemic” cause or exacerbate fear and panic among the public (Van Damme and Van Lerberghe, 2000; Bonneux and Van Damme, 2006; Barrett and Brown, 2008; Honigsbaum, 2009; Arafat et al., 2020; Maxmen, 2021). The WHO director Tedros Adhanom Ghebreyesus explicitly shared this concern in his press conference on 23 February 2020, where he observed that “using the word pandemic [now] does not fit the facts, but it may certainly cause fear” (WHO, 2020). Fear has an adaptive role in preventing people from harm, and researchers have previously noted that apprehensions about mass panic triggered by disasters and emergencies are exaggerated (Quarantelli, 2001; Gantt and Gantt, 2012; Van Bavel et al., 2020). In this light, the present research adds to the evidence base—instead of reacting with heightened anxiety, people in the 16 countries showed an increase in positive emotions and a decline in anxiety, and around the middle of March, anxiety ceased to be the predominant CE in all countries. This partially follows a natural decay in anxiety, as other studies have also reported (Mertens et al., 2023). However, results from time series analysis undertaken in this research provide strong evidence that the WHO's declaration of the pandemic was likely to have contributed to the decline in anxiety. It is well understood that uncertainty reduction in ambiguous situations lowers anxiety (Reuman et al., 2015; Chen et al., 2018) following evolutionary (Brosschot et al., 2016) and neurobiological (Grupe and Nitschke, 2013) pathways, and one possible explanation is that the WHO's declaration reduced uncertainty and confusion around the threat posed by the novel disease. Finally, this study also provided evidence that national lockdowns, declarations of states of emergency, and announcements of social restrictions did not result in a negative emotional reaction in any country. In none of the countries these measures coincided with an increase in anxiety or decline in positive emotions. On the contrary, Twitter posts in nine countries showed either a reduction in anxiety or an increase in positive emotions following such announcements, suggesting a largely positive reception of these measures in most countries. In this regard, the current study further emphasizes the need to interrogate assumptions around mass panic in the public sphere during periods of disaster and emergency.

Publicly available datasets were analyzed in this study. This data can be found here: https://zenodo.org/record/7213964.

AC, SS, and VB conceptualized the project and were involved in all the stages of the project including planning, data generation, cleaning, analysis, and preparation of the manuscript. AC and VB led the data analysis stage of the project. AC and SS led the writing of the manuscript. HS and JP led the data cleaning and LIWC processing of the dataset. AC led the project overall. All authors have approved the manuscript.