An outbreak of appreciation: A discursive analysis of tweets of gratitude expressed to the National Health Service at the outset of the COVID‐19 pandemic

Abstract Background The early stages of the coronavirus disease 2019 pandemic prompted unprecedented displays of gratitude to healthcare workers. In the United Kingdom, gratitude was a hotly debated topic in public discourse, catalysing compelling displays of civic togetherness but also attracting criticism for being an unhelpful distraction that authorized unrealistic expectations of healthcare workers. Expressions of thanks tend to be neglected as drivers of transformation, and yet, they are important indicators of qualities to which people attach significance. Objective This study aimed to use discursive analysis to explore how the National Health Service (NHS) was constructed in attention‐attracting tweets that expressed and/or discussed gratitude to the NHS. Methods Having determined that Twitter was the most active site for traffic relating to gratitude and the NHS, we established a corpus of 834 most‐liked tweets, purposively sampled from Twitter searches on a day‐by‐day basis over the period of the first lockdown in the United Kingdom (22 March–28 May 2020). We developed a typology for tweets engaging with gratitude as well as analysing what the NHS was thanked for. Results Our analysis, informed by a discursive psychology approach, found that the meanings attributed to gratitude were highly mobile and there were distinct patterns of activity. The NHS was predominantly—and sometimes idealistically—thanked for working, effort, saving and caring. Displays of gratitude were seen as incommensurable with failures of responsibility. The clap‐for‐carers campaign was a potent driver of affect, especially in the early parts of the lockdown. Conclusions The social value of gratitude is implicated in the re‐evaluation of the risks and rewards of healthcare and social care work in the wake of the pandemic. We caution against cynicism about gratitude overshadowing the well‐being effects that expressing and receiving gratitude can engender, particularly given concerns over the detrimental effects of the pandemic on mental health. Public Contribution This study involves the analysis of data provided by the public and published on social media.


| INTRODUCTION
The coronavirus disease 2019  pandemic is likely to present the highest collective level of biopsychosocial threat for a generation. While the early stages of the pandemic were characterized by mostly negative emotions, like fear and anxiety, 1 it will also be remembered in the United Kingdom for the conspicuous displays of gratitude to the National Health Service (NHS), healthcare workers and other key workers. Much has been written about the misalignment of gratitude rhetoric and material conditions for healthcare workers, from insufficient supplies of personal protective equipment (PPE) at the start of the pandemic, to problems with testing availability, to the much-derided 1% pay rise offer for NHS staff in England. 2 Twitter is a useful source of public data on expectations of healthcare. 6 As healthcare is reconfigured in the wake of the pandemic, the NHS needs to be recognized as an arena for cultural and social practices rather than merely an institution in need of organisational change. Paying attention to the qualities to which people attach significance could inform transformation through the lens of appreciation rather than criticism.
Gratitude as a personal, social and health benefit has grown in prominence during the pandemic. Consistent with past research that has shown that gratitude motivates prosocial behaviour, 7 studies focusing on COVID-19 found that participants who were grateful or thankful were more willing to endorse measures that helped curtail the spread of the virus. 8,9 Practising gratitude has been implicated as a predictor of well-being during lockdown. 10 Gratitude journaling is recommended in many of the online well-being courses that have proliferated during lockdowns, including the 'Science of Happiness' free online course that has attracted over 3 million enrolments. 11 There is a growing body of research that supports Fredrickson's 'broaden and build' model, in which experiencing gratitude, along with other positively valenced emotions, broadens the repertoires of action that people are prepared to take. 12 Immediate effects sparked by positive emotion tend to be relatively short-lived, but the model predicts that these actions build durable resources that can be drawn on as coping strategies to survive and thrive.
Our recent meta-narrative review of gratitude in healthcare identified a need for research on the ways in which gratitude acquires meaning in real-world situations. 13 Yoshimura and Berzins 14  This study is also an investigation into the potential for social media data to be explored using an approach informed by discursive psychology. It adds to the relatively few studies that have used this approach to explore tweets (e.g., Hurst 15 and Rasmussen 16 ). Unlike the everyday conversational routines that usually comprise the data source for discursive psychology, exchanges on Twitter are asynchronous and constrained by the features for interaction afforded by the platform. Tweets do, however, perform the types of social and psychological actions that are paradigmatic to discursive psychology, and the methodology has been proposed as having the potential to play a very important role in understanding online interactions. 17 Discursive psychology is a distinct branch of discourse analysis that looks at talk with respect to what it does rather than what it reflects. 18 The central concern of discursive psychology is how psychological characteristics are handled as part of participants' practices and orientations, performed as social action in how they talk 19 -or, in this case, how they tweet. This is not to discount the alignment of felt emotions with what people say, but a discursive psychology approach does not assume that what is said is a transparent relay to underlying states of mind. This approach gains support from recent experimental research in psychobiology that challenges notions of 'basic' natural emotions, arguing instead that emotions are 'made' and any category of emotion is filled with variety 20 -a stance that we consider to be particularly applicable to a complex emotion like gratitude.
Hitherto, the dominant paradigm for investigating the relationship between language and emotions has been cognitive psychology.
Cognitive approaches treat language as referring to or representing 'inner states': There is an assumption that there is a reality behind the talk that language allows us to access. 17 For example, Kleinberg et al. 21 assembled a 'ground truth data set' of emotional responses to the COVID-19 pandemic, arguing that the core aim of emotion detection is to 'make an inference about the author's emotional state'.
Rather than investigating whether a tweet is written in a pessimistic tone, they are interested in whether the author of the tweet actually felt pessimistic. While this aim is admirable, it is predicated on the questionable assumption that constructed texts are direct relays to people's emotions. In an excoriating critique of a study by Mitchell et al. 22 that used Twitter to map 'the geography of happiness', Jensen 23 has cogently outlined the dangers of conflating online social life with offline emotional states, along with other limitations of this type of research, such as sampling bias and overextending inferences.
To sum up, our research is couched in an ontology of the relational encounter: It starts from the point of view that gratitude takes place within the context of acts of communication. Underpinning our approach is a constructivist perspective, which assumes that social actors are in a continuous process of reinvention. Therefore, we do not assume a fixed identity for tweeters and eschew simplistic categorisations based on biographies. Here, gratitude is investigated as a discursive practice: It is strategized as purposeful, performative action, a social and cultural resource upon which actors draw. This approach rejects an epistemology of thanking expressions as the explicit representation of an implicit emotion and embraces one in which textual, verbal, visual and gestural actions participate in the pragmatic construction of the objects and objectives of language.

| Characterizing and compiling the data set
To inform our choice of the multiple available social media platforms on which to focus our research, we used the social media listening service Brand24 (brand24.com) to monitor traffic relating to gratitude and the NHS. Twitter was found to be by far the most active site for mentions of NHS AND (gratitude OR grateful OR thank OR thanks).
Twitter is a dynamic platform, and its affordances and features are often updated. At the time of data collection, Twitter is a publicly available platform that offers free accounts to those who wish to tweet or to follow (or subscribe to) specific accounts, but no account is needed to access or search the site. Tweet entry is limited to 280 characters. Tweeters can add up to four photographs, a graphic and short bursts of video. Twitter users can respond to a tweet by commenting on it, 'liking' it, quoting it and/or retweeting it. The visibility of tweets depends on the privacy settings selected by the tweeter, Twitter's proprietary algorithms that personalize what appears on users' news feeds and whether advertisers have paid for a tweet to be promoted.
A dictionary definition of gratitude is relatively uncontroversial: 'The quality or condition of being grateful; a warm sense of appreciation of kindness received, involving a feeling of goodwill towards the benefactor and a desire to do something in return'. 24 However, the nuanced meanings of gratitude, the appropriate application of the term and the characterisation of its value are highly contested. 25 Gratitude has been found to be prototypically organized: Features of gratitude do not belong to classically defined categories, the membership of which is specified by in/out criteria. Instead, gratitude is a concept made up of a 'fuzzy collection of features', some of which are considered more central than others. 26  Twitter is opaque about when it starts to impose its own filters or limits numbers of search results (its guidelines state that it filters for 'quality tweets and accounts'). 28  To construct the initial coding frame, we selected a stratified sample of 100 tweets from across the sampling period for inductive coding, leading to a list of characteristics useful for describing the tweet. This inductive approach is consistent with the 'emic' focus of discursive psychology, 17 in which we worked with the categories that we recognized in the corpus rather than imposing preconceived categories. We approached tweets as 'micronarratives' consisting of characters, actions, objects, contexts and instruments. 30 Informed by Haugh's study of im/politeness as social practice, 31

| Coding the data
For the full data set of 834 tweets, we followed the principles and protocols for consensual qualitative research-modified. 32,33 Coding was additive: Each tweet was coded for at least one function and one plot, but as many codes as were relevant were applied. GD coded all the tweets, GR coded 60% and KLG coded 40% so that each tweet was coded independently by two coders before they were discussed and coding agreed. AMR audited the coding. GD and KLG narratively coded for metaphors in tweets. Additionally, in vivo coding was used to capture explicit mentions of what the NHS was being thanked for and references to groups or individuals to whom the thanks was being addressed.

| Ethical considerations
Although Twitter is a public platform and this study does not include sensitive personal information, private people may have an expectation that their tweets are specific to the context of Twitter rather than being the subject of research. We have drawn on recommended frameworks for ethical use of social media in research. 34,35 Examples of tweets reported verbatim in our analysis are from corporate accounts or public figures for whom there is a reasonable expectation of publicity, 36 or we have obtained explicit permission to quote the tweet. Examples for which it has not been possible to obtain permission have been paraphrased.

| RESULTS
Most of the expressions of thanks to the NHS in our data set were 'behabitives', in that they enacted the social behaviour of thanking by their very expression. 37 These expressions were often implicated in a variety of other functions, of which the most prominent were sharing news, describing care experiences, giving instructions or making requests and commenting, critiquing and criticizing. Gratitude was also harnessed to narratives of generosity through offering or receiving benefits (such as donations of goods and discounts) and fundraising as a material form of gratitude. Figure 1 shows a thematic analysis of the free coding of text in tweets that were specific about what the NHS was being thanked for. We have aggregated personal attributes for which people were thanked under 'virtues': these were dominated by dedication, selflessness, kindness and bravery, but commitment, courage, generosity, positivity and compassion were also mentioned.
A frequency analysis over time ( Figure S1) shows that the number of tweets expressing gratitude to the NHS ramped up in the days preceding lockdown. For the next 5 weeks, a cyclical pattern of peaks is evident, showing that the social movement campaign, clapfor-carers, on Thursday evenings served as a potent attractor for tweets of gratitude over this period. Although there was no one turning point at which gratitude to the NHS became less visible in our data set, by the end of April, criticisms of clap-for-carers were beginning to take effect and the event started to lose traction. This is consistent with the findings of McKay et al., 38 who, in their analysis of tweets associated with the NHS and COVID during the first lockdown, found a decrease in engagement after the first month, which they attribute to lockdown fatigue and the effects on tweeting habits of the limiting experience of staying at home.

| The clap-for-carers effect
Clap-for-carers, or more properly, Clap-for-our-carers, was a UKwide campaign that encouraged people to take to their doorsteps, balconies and windows to give a round of applause to NHS workers and other key workers every Thursday night at 8 PM. It originated from an Instagram post from Dutch Londoner, Annemarie Plas. She was inspired by videos posted on social media of hospital healthcare workers being applauded in Italy in mid-March 2020, a practice that spread quickly around Europe. 39  Tweets associated with clap-for-carers tended to include performances, with videos of applause often shared in these tweets.
Tweeters also used it as an opportunity for words of appreciation directed to the NHS and key workers-numbers of tweets coded for 'recognising', 'performance' and 'words of appreciation' peaked on Thursdays throughout the study period.

| The idealisation of work
Over a quarter of tweets thanking the NHS implicated work in their appreciation. The most often used qualifier was 'hard', but references were often made to time: 'round the clock', '24/7', 'day and night'.
The adverb 'tirelessly' was most often associated with 'working'. A typical example was: We want to thank every person who's working tirelessly to keep this country healthy. We're so lucky to have the NHS and want you to know how grateful we are for your selfless hard work during this terrible time. We're #StayingAtHome for you and the incredible work you're doing (3 April, @jlsofficial).
There is pronounced asymmetry-perhaps even an ironybetween the characterization of work in thankful expressions and the nature of the work being thanked for: Someone can be praised for 'doing fantastic work', while the rationale for thanking is predicated on an acknowledgement that the work was not fantastic to do. Similarly, describing NHS staff as working 'tirelessly' is contradicted by the welter of narratives of strain that emphasized fatigue and exhaustion. In common with McKay et al., 38 we were struck by the powerful disconnect between the symbolic and the tangible that emerged in tweets relating to the pandemic. They found that symbolism centred on the language and performance of valorisation was undercut by escalating hospitalisations and deaths. This disconnect is especially evident in our data set in the ways in which 'saving' was referred to as a reason to thank the NHS.

| Caring made visible
In her wide-ranging exploration of care published at the start of the pandemic, Bunting 43 describes care as the 'invisible heart' and calls for greater acknowledgement in terms of recognition, funding, respect and DAY ET AL.
| 157 value. The provision of care extends to a much wider context than that provided in the NHS to which our analysis specifically speaks. However, the prominence of care as an object for people's thanks suggests that care became more visible, and better appreciated, in pandemic discourse. The word 'care', unlike 'work', connotes a relationship, reinforced by the verbs with which it was often accompanied in tweeter's phrases: 'giving care' and 'taking care'. Gratitude for care tended to be more specific than those for 'work' or 'saving', with about half referring to treatment experiences of named patients. The prominence of narratives of care was also apparent in our analysis of those to whom gratitude was addressed. Carers were the third most-mentioned thanked category, after workers and staff. The word 'carers' in 'clapfor-our-carers' was probably chosen for reasons of alliteration, but, given the early-stage success of the campaign, it may have contributed to raising the profile of all carers. An outcome may be that the social reimagining of care work done by those exposed to risk and precarity, as called for by Rossiter and Godderis, 44 becomes more likely in a postpandemic world.

| Meaningless or meaningful?
Sorace has argued that gratitude is the 'ideology of sovereignty in a crisis', too easily slipping from the recognition of individuals to an acceptance of the systems that reproduce their exploitation. 45 When gratitude was suspected as being used as a substitute currencysupposed to compensate for low pay and unsafe working conditions, or to offset policies deemed exploitative-it was reacted to, unsurprisingly, with suspicion and resentment: Emotions voiced by tweeters that were often allied to expressing gratitude, particularly in association with clap-for-carers, were pride, love and hope. In our data set, about one in six tweets invoked solidarity and 'togetherness' as a value that they appreciated.

| DISCUSSION
This study set out to explore gratitude expressed in tweets to the NHS during the early part of the pandemic. We adopted a methodological approach that moved away from the dominant sociocognitive model for investigating gratitude to one that is more discursive. We followed the stance of those exploring other situated social actions, such as apologies, 46 49 Greenberg et al. 50

| Implications
Our study supports the contention by Shaw 51 that gratitude is implicated in assurances of 'mattering' that contribute to a moral community. Benefits that stimulate gratitude convey the notion that others care about us and that we are worthy of their care. There is an evidential base for gratitude being linked to social, emotional and psychological well-being. 52 In the face of a concomitant mental health pandemic, the denigration of acts of gratitude, on social and in mainstream media, may discourage people who could potentially benefit from the well-being effects of practising gratitude. Previous research has concluded that people value opportunities to express their thanks. 4 A recent House of Commons report highlighting the impact of workforce burnout identified lack of recognition as a significant contributor to feelings of 'abandonment' from sectors, like social care and pharmacy, that felt excluded from the public recognition being afforded the NHS in the early part of the pandemic. 53 As we 'build back better', attention needs to be paid to spaces and places that accommodate gratitude-not only in healthcare but also in society in general. This is not to say that expressions of gratitude should be immune to criticism, but people's anticipation of accusations of inauthenticity and virtue signalling may discourage thanking activities that, if enacted, could make a real difference to motivation and morale.
Our study also has implications for how the pandemic is remembered in popular culture and commemorated. Clapping has already become a shorthand for social appreciation, although sometimes with ironic overtones. We need look only to the way in which sentiments like 'Blitz spirit' still influence people's strategies for coping in times of crisis 54 to realize that the way the pandemic is commemorated will influence how we respond to future crises. Given how contentious thanking the NHS became during the pandemic, the ways in which gratitude to healthcare workers is incorporated into commemorative acts and material culture should be the subject of extensive consultation to maximize its chances of striking the appropriate tone.

| Strengths and limitations
An innovative aspect of this study is that we have developed a typology for thanking expressions that may be applicable to categorizing gratitude in other contexts. This study benefited from a robust approach to data collection and analysis. Tweets were considered as a whole-including images and videos-which are not usually captured by data-scraping methods. The consensual approach to coding, while time consuming, allowed for a reflexive attitude to our data.
We do not claim, however, that tweets constitute the 'naturally occurring talk' preferred by those using a discursive psychology methodology. Although they are 'natural' in that they are not intentionally solicited by a researcher, the search-and-retrieval methods necessary to assemble a data set, and the opacity of Twitter's proprietary search algorithms, make data retrieval analogous to elicitation.
Acts of true creativity in thanking practices are likely to employ semantics that elude search strings even when those acts are highly culturally salient. An example is that the discourse surrounding Captain Tom's 55 extraordinary fundraising activities for NHS Charities did not feature strongly in our data set in spite of being a dominant narrative that featured an outpouring of gratitude during the first DAY ET AL.
| 159 lockdown. Thanking exchanges took place mainly between Captain Tom and donors to his campaign, with the NHS being invoked only occasionally in tweets that fulfilled our inclusion criteria. By focusing on 'micronarratives', some of the 'macronarratives' may be underrepresented, both because of our restricted search terms and by the purposive selection of attention-garnering tweets rather than relying on a random sample. As Venditti et al. 30 have pointed out, social media use is driven by more than the spontaneous practices of users-it includes strategized activities, including 'liking', that are determined by the specific architecture of the platform. Insights into the pragmatics of user interactions are not available to researchers examining content, and care needs to be taken not to equate 'liked' tweets with public approval beyond the context of Twitter.

| CONCLUSION
We have presented a framework for analysing gratitude expressed on social media that we applied to attention-attracting tweets that en-