Membership categorisation and antagonistic Twitter formulations

Introduction

The rapid and widespread uptake of social media platforms such as Twitter, Facebook, Flickr, Instagram and YouTube has created new ways for people to interact and share information. This brings both benefits and risks for civil society and new challenges for agencies responsible for ensuring that the boundaries of acceptable and legal behaviour are not crossed and, if they are, that appropriate action is taken (Webb et al., 2016). In this respect, the proliferation of so-called antagonistic speech ¹ in social media is an area of growing concern, as recent high-profile examples confirm. For the social sciences, the migration of so-called hate speech to social media platforms affords new opportunities to study antagonistic behaviour(s), understand the impact of social media and identify ways in which government agencies and civil society organisations might respond more effectively to its threats and consequences. It also brings into sharp focus debates surrounding freedom of speech and the regulation of open, networked and distributed communications in the digital age. At the same time, we acknowledge that ‘hate speech’ is a normative concept and recognise the ethnomethodological requirement to respecify (Lynch and Bogen, 1996) this term through the study of practical action. Consequently, in this article we focus on the membership category features of high-profile tweets that have been identified, and possibly designed, to act as ‘ignition points’ for the generation of antagonistic communication flows – a potential resource for generating interest, attention, economic value ² and the dissemination of particular types of ‘world view’.

Twitter has been described as a new digital agora wherein debate and the exchange of views can facilitate the public and civil sphere in ways that promote democratic engagement and communicative rationality (Edwards et al., 2013; Housley et al., 2014). However, Twitter has also been used to promote misinformation and antagonistic content (Webb et al., 2015). At the same time, Twitter has been a popular platform through which social scientists have been able to engage with big and broad social data due to its accessibility and open access application programming interface (API) that has enabled researchers to capture data and interrogate massive data sets with a range of digital tools, analytical techniques and interpretive procedures (Housley et al., 2014). However, to date little work has been carried out on the social organisational characteristics of Twitter interaction (Giles et al., 2015). This is of importance if we are to fully grasp the pragmatic characteristics of various phenomena such as public reaction to civil disturbance, the role of rumour and misinformation online and provocative or discriminatory speech acts. For analytical purposes, we understand these everyday speech practices as antagonistic due to the way in which they, as a consequence of their design features, elicit oppositional and relational responses.

One promising avenue for exploring this form of online behaviour is the analysis of thread interaction on social media (see Housley et al., 2017; Zubiaga et al., 2016). Drawing on conversation analysis (Sacks et al., 1974), this approach potentially provides insight into the categorical and sequential characteristics of online, networked and multi-party interactions. However, there is also a variety of technical and ethical limitations to the use of Twitter data in this way that revolve around issues of consent that are yet to be fully resolved. ³ Despite these ethical dilemmas, the study of high-profile public social media accounts remains of analytic interest and less problematic in terms of emerging ethical frameworks for the analysis of social media.

During the course of this article we examine a selection of tweets gathered from a high-profile celebrity account that has become notorious for its capacity to generate antagonism. The account is framed in terms of the celebrity ‘Katie Hopkins’, who is also the author of a popular UK newspaper column and has made several reality TV appearances in the United Kingdom. Utilising the Twitter streaming API, we were able to harvest over 12,000 tweets posted by this account, and collate and classify them in terms of topic and a range of network metrics (e.g. retweets and the number of responses posted to each tweet). In this article we explore a selection of tweets from this corpus and order them in terms of topic and number of replying tweets where appropriate. We use the number of replies as an interpretive lens in order to explore potential design features associated with different types of ‘tweet formulation’ generated from this account – where the number of replies may act as a signature of antagonism and controversy. Current and future work is examining the character of these replies as forms of counter-speech and other forms of social media ‘thread interaction’. However, in this article we focus on tweets generated from a high-profile public account as accounts and, more specifically, membership categorisation devices (MCDs) that might be understood as ignition points for significant socially mediated digital antagonism. In this sense, this article focuses on the MCD features of potential points of origin for antagonism within open social media environments such as those supported by the Twitter platform.

Ethnomethodology and the pragmatic features of social media interaction

MCA and media

In terms of membership categorisation and accounts within media settings, a range of studies have examined how moral reasoning is mobilised in order to generate debate within a variety of settings. These include current affairs radio phone-ins (Hester and Fitzgerald, 1999) broadcast news (Fitzgerald and Housley, 2016), congressional hearings (Lynch and Bogen, 1996) and televised party political debates (Fitzgerald and Housley, 2016). In addition to these formats, studies have included more traditional textual formats such as letters to the editor in major national and international newspapers, and the construction and category-based design of headlines (Lee, 1984).

The study of headlines as category-based sense-making phenomena represents a relevant example for exploring how membership categorisation practices are related to social media communications. In an age of networked digital platforms where people are urged ‘to broadcast themselves’ within 140 characters, the newspaper headline makes for an interesting comparative artefact from an earlier form of media ecology and environment. There are important similarities and differences; the differences relate to the emergence of disruptive digital communication networks, where social media can be viewed as flattening earlier modes of media broadcast organisation where the few speak to the many (Edwards et al., 2013). For example, Dutton (2009) and (Newman et al., 2012) have argued that social media platforms have contributed to the emergence of the ‘Fifth Estate’, thereby allowing audiences to share their views with worldwide publics and to engage directly with one another in this new public space and with international news media organisations.

The similarities include the recipient configured design demands of brevity and associated considerations that relate to practical communicative orientations towards audiences and potential readers. They also include the use of membership category work and related conversational practices that include ‘extreme case formulations’ (Pomerantz, 1986), contrast devices (Atkinson, 1984) and other forms of mundane, rhetorical order. A key idea here is the way in which membership category work can be understood as types of formulation that morally account and index social activities, associations and attributions. In an early lecture and set of notes, Sacks (1992b: 170–199) discusses a newspaper article in detail, which is headlined ‘A Navy Pilot Calls Vietnam Duty Peak of Career’. In this piece Sacks explores membership categorisation practices as particular types of moral formulation that are central to understanding the social organisation of moral accountability; for example, in terms of making sense of being shot at, responding to attack and bombing, and how this is categorised and formulated via print media to a public audience in relation to a controversial topic. For Sacks, a central issue here is the operationalisation of the descriptive cultural apparatus of membership categorisation and the deployment of MCDs within accountable settings such as print media. Sacks states (1992a),

It may be much noted that the choice of the device is not just to be made by reference to their consequences or the like, but by reference to how, given the use of the consistency rule to formulate his alter egos, how it provides for the formulation of actions. The availability then of making his categorisation decision in such a way that it routinely provides for a categorisation of his opponents, and by virtue of the mutual categorization then an assessment of either’s actions is a crucial matter. (p. 206)

In terms of social media interaction, we mobilise the notion of membership categorisation formulations as condensed accounts (Housley and Fitzgerald, 2009) that can be understood as an everyday method for advancing, framing and contesting claims through the mobilisation of the ‘descriptive cultural apparatus’ identified by Sacks (1992a) and advanced by others (Fitzgerald and Housley, 2015). These account formulations are relevant in two ways. First, the economy and space afforded to Twitter posts (140 characters) lend themselves to an economy that is different (but similar) to the ‘strict meter’ characteristics of other media communications such as newspaper headlines (Hester and Fitzgerald, 1999; Lee, 1984). Second, the interactional milieu of antagonistic communication are predicated upon claims and counter-claims that are built around matters of identity, difference and their category-bound associations, attributes and activities (or predicates; see Eglin and Hester, 1992). However, this work also occurs within an environment that consists of functional actions that are built into the social media platform in question. These afford certain ‘sequential’ and networked characteristics; while not a focus of this article, they are nevertheless an important feature of antagonistic multi-party interaction on Twitter and similar forms of social media.

Twitter and antagonistic account formulations?

Tweets can be understood to perform a range of actions that are enhanced through specific technical and culturally bound procedures. For example, the use of links, hashtags and the ‘@’ to particular accounts within real-time posting are integral to the network dynamics and ‘information flow’ of Twitter-based communications. Furthermore, specific roles within the network hierarchy and social media ecology afford and reflexively define how one is read and heard.

One strategy for accomplishing interest and online ‘status’ is through the performance of antagonism. These performances can be occasioned, as part and parcel of the speech act moves associated with the bob and weave of interacting online in relation to topics of interest – from politics, sport, entertainment to religion and everyday troubles. Or, they may form a key aspect of the performative apparatus through which individuals or groups present themselves within open and public social media networks such as Twitter.

The rise of the so-called ‘celebrity troll’ (McCosker, 2014) and online cultural commentator represents one such popularly encountered form of network role where antagonistic formulations form the bread and butter of the everyday online work and task of generating interest, network capital and attention within a wider media ecology. It may also be used as a vehicle for using controversy as a means of broadcasting specific world views; the migration of these activities from traditional media to social media stands outside the scope of this article. However, the transformative power of Web 2.0 and the interactivity generated by social media platforms transforms these forms of activity and generates spaces for innovation that may have wider consequences for the digital agora and civil conduct. As stated earlier, this article contributes to that work by examining the formulation of high-profile ‘ignition’ tweets that play a crucial role in generating antagonistic content.

Context and data set

Using the Twitter search API, we collected a total of 12,314 tweets posted by Katie Hopkins from her @KTHopkins account over the period from May 2009 to May 2015. Following this, we then used the Twitter search API to collect all the replies to each of these ‘source’ tweets. The source tweets and their replies were then visualised as conversational ‘threads’ with the source tweet at the top and subsequent replies underneath. In most instances the interactions occurring only took the form of replies to the source tweet. Katie Hopkins rarely returned to post in reply to any of the responses she received. In some cases users responded to each other, so sub-conversations took place within the thread.

The threads were collated into different categories according to the number of responses generated by the opening post. Although this was not an intention at the outset of the analysis, this provided a useful metric and enabled us to consider how the alternate construction of tweets might relate to the number of responses it generated.

OPEN IN VIEWER

The threads were reviewed by members of the research team to inductively identify areas of interest. A sample was then selected for analysis. Once the topic of antagonistic Twitter formulations had been chosen as a topic of study, the data set was reviewed and threads of interest were collated. We grouped threads together according to the topic of the original post – working mothers, ginger hair, teachers’ strike and obesity. Each group contained threads from the 11–100 replies, 1–10 replies and 0 replies subsets. The opening post of each thread was analysed to identify interactional features occurring and their relation to the construction of antagonistic formulations. We then looked across the selected data to identify patterns regarding the use of membership categorisation practices. The posts discussed here have been selected to represent the overall analytic findings.

Building posts as tweet formulations

We begin the analysis with some general observations about how posts by Katie Hopkins are built as tweets. That is, the ways in which the design of posts relates to the features and affordances of the Twitter platform in ways that draw from everyday methods and conversational resources. These formulations draw on Twitter’s status as an open platform that enables the broadcasting of pronouncements or questions to a general audience. This can be achieved through reference to particular ‘@’ function ‘handles’ that act to mark individual user(s), the use of generic pro-terms such as ‘you’ (e.g. as observed in Tweets 1 and 3) and the use of hashtags that topicalise posts in ways that allow for computationally driven ‘discoverability’ within networked social media streams. In one sense hashtags provide a self-generated topical index and can also serve to make a further comment or moral assessment available to a general audience (e.g. the hashtag operationalised and included in Tweet 1). A final feature here is the limited word count; this often leads to deviation from standard grammatical frames, a concise style and the absence of opinion markers (Tweets 2 and 3).

OPEN IN VIEWER

Tweet 1

(123 responses).

OPEN IN VIEWER

Tweet 2

(5 responses).

OPEN IN VIEWER

Tweet 3

(23 responses).

The tweets identified above reference a range of topics. At a very general level, we might conclude that each one (Tweets 1–3) is emotive and expresses a strong opinion that is likely to inflame others. However, in this article we argue that these formulations have specific features and methods of design that are oriented towards generating antagonism and response. In the following sections, we look at them in further detail to identify how these formulations are designed to be antagonistic, and note several different kinds of membership categorisation practices that achieve this.

Membership categorisation practices and the design of antagonistic tweets

Here we discuss three types of membership categorisation practices that are routinely found in the data set and appear to be central to the design of posts as antagonistic formulations that are reflexively constituted in an through the affordances of the Twitter platform. We discuss each type in relation to a particular topic. Within these routine practices we can begin to observe some differences that may have bearing on the number of responses posted to an opening tweet, as evidenced by the data corpus.

Category use and feature design

In the examples above, the tweets make reference to the topic of obesity; in both cases the membership categorisation device ‘fat’ (as an ‘n’ population membership category device) can be seen to be in operation. The first formulation (Tweet 4) operationalises a form of contrast predication (Housley and Smith, 2015) between not being ‘fat and happy’ on the one hand and being fat and having ‘no will power’ on the other. The use of contrast classes within political rhetoric and political texts is well documented (Atkinson, 1984; Hester and Fitzgerald, 1999). The second formulation, presented as a second example here (Tweet 5) within this range of responses, is designed differently.

OPEN IN VIEWER

Tweet 4

(123 responses).

OPEN IN VIEWER

Tweet 5

(95 responses).

In this case, the category incumbency of ‘being fat’ is problematised through the assertion that ‘there is simply no excuse for being fat’ (Tweet 5); the opening part of the formulation is followed to be a framing of the ‘condensed account’ in terms of humour: ‘… the funny things is – fat people always have an excuse’. Finally, possible excuses are listed in parenthesis – they can be understood to form a collection of category predicates that signify a particular reading of obesity, in this case mundane biological or morphological characteristics. In the first formulation, the use of contrastive sets of predicates makes use of standard practical methods for generating debate or controversy, while the second formulation invites responses by suggesting that while there are no excuses for being ‘fat’, members of this population device routinely provide excuses of the sort that could be added to the list. The list device provides a further resource that invites others to add to the collection of predicates provided and in ways that remain consistent with the particular moral framing of the relevant membership categorisation device and topic being discussed.

In this example, the formulation is organised in terms of ‘issues’ for the ‘over eater’ and ‘the taxpayer’. The formulation is organised in three parts: the first part frames the topic of obesity (that can also be understood as a membership categorisation device in this and other examples) in terms of the predicate (or category-bound attribute) of ‘issues’. Issues such as knowledge are understood to be owned by specific membership categories, in this case ‘the taxpayer’ and ‘over eater’. Once again this sets up a contrast class, here between the membership categories in question, that is central to generating debate. However, this contrast is not left open – the final part of the formulation introduces a third membership category, in this case an institutional category, namely the National Health Service (NHS) that is mapped to a predicate of making payment by an open and omnirelevant membership category of ‘someone else’.

This formulation (and others like it) received fewer responses than the examples discussed earlier (Tweets 4 and 5). It goes without saying that this remains a qualitative signature and does not constitute some form of quantitative feature identification. However, in terms of this analysis we might note that the three-part structure of example Tweet 6 resolves the contrast (and thence provocative force) between ‘taxpayers’ and ‘overeaters’ set up within the first two stages of the tweet formulation that may detract from its antagonistic potential.

OPEN IN VIEWER

Tweet 6

(5 responses).

Moral work

Within the following set of examples, industrial action by teachers in the United Kingdom is topicalised and broadcast. In this section we explore how membership category use and feature design is put to use as a moral apparatus for downgrading and framing particular groups and activities.

The first example (Tweet 7) frames the topic and setting ‘Teachers on Strike’; the formulation mobilises the institutional pronoun ‘We’ and deploys the first pair part of a moral relational pair, namely blame (Housley, 2002) in relation to ‘the government’, and the collective pronoun, in this case, of ‘You’. This account is a type of blame allocation formulation, which is concluded through an identification of moral expectation and predication, namely that if you want to be a teacher you ‘Get on with it’. The second example (Tweet 8) draws on similar membership categorisation resources in order to generate moral controversy and formulate particular social groups, agencies and institutions in particular ways through morally charged forms of association, attribution and activities that we understand as category predication (or category bounded) work. In this way, contestable and antagonistic world views are generated which categorise particular social groups and events in morally charged ways, often through sharp contrasts or extreme case formulations based on mundane senses of moral proportionality, frequency and volume.

OPEN IN VIEWER

Tweet 7

(54 responses).

OPEN IN VIEWER

Tweet 8

(44 responses).

In Tweet 9, identified from a number of tweets that generated between 11 and 100 responses, the topic of the ‘Teachers’ strike’ is populated with a number of membership categories that consist of ‘educators’, ‘babysitters’, ‘child’, ‘school’ and ‘working mums’. A moral relational expectation is established: ‘you expect your child to be at school on a school day’ – children and being in a school on a school day is presented as routine, normal and expectable. However, strikes disrupt this membership categorisation formulation whereby ‘working mothers’ are penalised by strike action and by the group pursuing such forms of action. However, ‘teachers’ are not directly referred to even though they are relevant to this moral account and in ways that can be heard to position them as responsible and morally identifiable as a professional group that is pursuing forms of action that ‘penalise working mothers’.

OPEN IN VIEWER

Tweet 9

(23 responses).

Instead, reference is made to ‘Educators’, a membership category term that is more general and less specific as a group. Consequently, the clarity of membership categorisation and formulation strategies such as blame allocation, use of contrast classes or forms of predication and ‘us’ versus ‘them’ rhetoric is less discernible in this formulation. It is membership category rich and may be open to a range of imputations in comparison to other forms of antagonistic formulation design within the economy of Twitter’s 140 characters.

Accountable social action and sense making

The penultimate set of examples identified from the corpus relate to mothers and work. Of interest here is the significance of this social group in terms of gender, rights and discrimination and the way in which it resonates with Sacks’ initial explanatory formulation for membership categorisation as a method for accountable social action and sense making. In these examples, membership categorisation work does not merely provide for the moral evaluation of actions and specific groups, but – in a more recognisably explicit format – calls specific social groups as membership categories not only to account, but into question as viable everyday ontological positions.

In Tweet 10, the formulation operationalises the membership category ‘Mummy’ – a membership category of the device ‘Family’ although it can also be understood to act as a device in its own right. The membership category is tied to a further device, namely work, through reference to the category-bound activity of being ‘full time’ and the associated membership category of ‘occupation’.

OPEN IN VIEWER

Tweet 10

(115 responses).

In this formulation, the work of being a full-time mother, that might involve childcare, emotional labour and domestic activity, is not only downgraded but refuted; it is deemed ‘not an occupation’. In this sense, the membership category of ‘Mummy’ is degraded and any index to work as a morally valued device with associated (potentially positive) imputations and inferences is dismissed. The economy of this formulation, and the use of a contrast class device, is realised in the second part of this formulation – ‘It is merely a biological status’. Of note here is the use of ‘merely’ as an extreme case formulation (Pomerantz, 1986) and the severing of the membership category of ‘Mummy’ from any category-bound connection with ‘work’ through an exclusive (and reductive) connection to a ‘biological status’.

The second example (Tweet 11) identified from those tweets that generated the highest range of responses (1–100) within this topic domain exhibits similar design features. However, rather than proliferate or contrast potential imputations through the use of the device of ‘biology’, the formulation works consistently to downgrade and degrade the membership category of ‘Mummy’ being tied to any category-bound features associated with work and its associated moral and cultural value.

OPEN IN VIEWER

Tweet 11

(30 responses).

The third example (Tweet 12) gathered from the corpus concerned with work and mothers is formulated as a two-part moral directive where predicates such as ‘purpose’, ‘women’ and ‘work’ are tied together. The practice of childcare (a possible category-bound feature of the membership category ‘Mummy’ and the wider device of ‘Family’) is relationally bound to the practice of ‘investment in future earning potential’ – a type of activity and practice that we can hear to be bounded to the device ‘work’. This formulation can be understood to make two principle moves, first by associating ‘women with work-as-purpose’ and second by upgrading the practice of childcare in relation to economic security and career advancement. As a directive, it does not explicitly make contrast or downgrade other ways through which being a woman or mother might be formulated. However, it does advance a particular morally constituted world view that is contestable.

OPEN IN VIEWER

Tweet 12

(5 responses).

Mobilising response

The analysis so far has revealed what kinds of features build a Twitter post as antagonistic. It underlines that the design of these posts does not just relate to the inclusion of key words that can be widely understood as offensive or inflammatory, or to the inclusion of controversial topics. Instead, we can observe more nuanced practices which draw on the affordances of the Twitter platform, deploy particular categories, perform moral work, and call specific groups into account and challenge their viability. Through these observations we can understand how posts may be built in such a way that they provoke a response from others. A particular feature of our data collection approach is that we can observe the number of responses generated by each opening tweet. By comparing across the data set, we can note that there are often substantial, even systematic, differences between tweets that receive multiple responses and those that receive very few. Although we cannot comment on or measure causation directly, this metric allows us to make some further observations about the varying extent to which posts made by Katie Hopkins mobilise response from other Twitter users. Tweets 13–15 all received fewer than five open interactive responses via the ‘@’ function.

OPEN IN VIEWER

Tweet 13

(0 responses).

OPEN IN VIEWER

Tweet 14

(0 responses).

OPEN IN VIEWER

Tweet 15

(5 responses).

Tweet 13 is another example from the obesity topic collection. In this formulation, the condensed account is prefaced by an unsubstantiated statistic. A directive is then provided: ‘need to privatise our health service’ and ‘increase the premiums’ for ‘people who cannot control greed’. In this formulation, a general population device (namely, people) is deployed and mapped to the category-bound attribute ‘for people who cannot control greed’. More explicit and, we might suggest, provocative membership categories such as ‘fat’ are not deployed. Again, we are not aiming to make any concrete causal observations here, but note that there may be a relationship between the design of the formulation and the interest and the ‘thread response’ generated.

Tweet 14 comes from the topic of the UK teachers’ strike. It exhibits similar characteristics to tweets in the same topic, as described earlier. However, in terms of category proliferation, the first part of the formulation predicates the MCD of ‘man’ in five principal ways. The second part of the formulation consists of a contrast class between the private sector and public sector that makes use of a gear metaphor as a contrast device. The gear metaphor is formulated in terms of an overarching descriptive apparatus that is described as ‘the five speeds of man’; this serves to generate a list of activities that serve to situate the act of ‘striking’ as the most static member of a collection of mobility categories. Of note is the relatively complex design of the formulation in this case; however, it is a ‘weaker’ form of category formulation, in the sense that the public and the private are primarily contrasted through modes of predication (such as evolving or regressing) that may not be heard to be as recognisably provocative as possible alternatives. Furthermore, the membership category predicate proliferation that is operationalised through the gear metaphor and predicate list set up multiple imputations that may detract from the economy of antagonism contained within different tweet formulations, restricted as they are, by 140 characters.

Tweet 15 comes from the working mothers’ topic domain within our data corpus and is mobilised as a story formulation. A fictional mother and her two (male) children are presented at the preface. Crucially, Emma is a ‘full time mum’; no other reference is made to other family members, husbands or partners. Although it is not explicitly referenced, we might potentially infer that ‘Emma’ is a single parent – a membership group historically and routinely subject to moral censure. The second and third parts of the story formulation speculate on ‘what would happen’ if they occupied the membership categories of ‘part-time mum’ or a ‘flexi-time mum’ – forms of membership categorisation that can be tied to the membership categorisation device of work. This opens up a range of potential inferences. However, in comparison to Tweets 10–12 it does not refute any index with work; rather, the device of time (and associated category-bound features such as effort or value) presents an additional organising principle that has potential imputations that may or may not be morally degrading. As a consequence, we might suggest that this formulation and membership category design is less antagonistic in comparison to other examples discussed earlier.

We can also make a more direct comparison between two very similar tweets (Tweets 16 and 17), noting the changes made between the one posted first and the one posted second. These differences could account (in part) for the varying numbers of responses. These tweets relate to a new topic – ginger haired children.

OPEN IN VIEWER

Tweet 16

(14 responses).

OPEN IN VIEWER

Tweet 17

(181 responses).

The first formulation (Tweet 16) is subject to a considerable number of responses. The reference to ‘Babies’ can be heard to act as a membership categorisation device in this instance as well as a membership category of the wider device of ‘Family’; in this sense we can hear it as being duplicatively organised (Watson, 2015). In this formulation the membership category ‘babies’ is predicated in terms of hair colour, in this case ‘ginger’; this provides the grounds for generating an additional membership category population of ‘ginger babies’. The second part of the formulation invokes the consistency rule in relation to the membership category in question, but stipulates an exceptional category-bound (predicate-based) feature; namely that they are ‘so much harder to love’. This formulation makes use of a contrast class (regular babies vs ginger haired babies) and forms of moral downgrading, as well as an extreme case formulation through which to degrade a population group and, we might suggest, antagonise, provoke and generate a response.

The second formulation (Tweet 17) is one that was produced by the account in question later in the timeline of social media postings gathered within the data set. Of interest here is the difference in the number of responses that may be tied to the design of the MCD formulation. First, the target membership category is referenced in a more economical manner, that is, ‘ginger babies’ (as opposed to ginger haired babies) followed by a second part reference to ‘Like a baby’, which establishes a possible hearing of the first part as a separate membership category device in its own right – rather than being merely a derogatory description within a moral hierarchy of membership categories associated with ‘regular babies’. Furthermore, the formulation is broken down, through the use of punctuation, into a clear three-part list, a well-established rhetorical device (Atkinson, 1984). We suggest that this combination of design features – that is, contrast, use of a three-part list, economy and the establishment of a distinct ‘othering’ device – may constitute a powerful antagonistic design that is optimally refined in comparison to the example provided by Tweet 16.

Conclusion

During the course of this article we have described how tweets can be crafted to draw out responses from others. Our aim has been to examine how membership categorisation practices are central to the design and configuration of tweets as ‘ignition points’. While we are not concerned with establishing a clear causal link between antagonistic formulation design and the number of responses to different social media posts, we suggest that this form of thick MCD description of formulations provides the grounds for carrying out further work and analytic investigation of the characteristics, content and category organisation of content that provokes and antagonises sections of the online population, often through the targeting of specific membership categories and the promotion of morally charged category-bound evaluations.

In the age of Web 2.0, social media and distributed networked ‘public’ platforms, it is important to trace continuities and discontinuities between old and new forms of communication that are oriented to promoting particular world views and the moral evaluation of particular social groups. The potential to build up network capital through routine antagonism and agitation is a well-established form of popular political activity, as evidenced by the study of traditional media. In the digital age it is imperative to understand how high-profile communications and formats (such as widely broadcast tweets about particular social groups) are being repackaged for existing and new audiences through new forms of communication. Furthermore, an understanding of how interest, debate and antagonism are generated by key public agents enables a more nuanced understanding of the emerging digital landscape and its relationship with popular political perceptions. To this extent, the analysis of antagonistic membership categorisation formulations provides insight into the practical and moral organisation of controversy and the socio-technical generation of antagonism in contemporary times.