2. Data generation

As a convenience sample, a first-year Language and Communication class at Cardiff University (Wales, UK) was asked towards the end of one lecture to retrieve from their phones five messages that they had either sent or received in the previous week and to transcribe them as accurately as possible (i.e. 'exactly as they appeared on the display screen'). This was done at the end of a questionnaire study conducted by Brown (2002) which also examined patterns of SMS use among the students (e.g. reasons for using it, people they sent messages to, and whether or not they used 'predictive text') and other practical considerations such as the amount of money spent on text-messaging, the person who pays the bills, and the network used. (For similar surveys see also Grinter & Eldridge's (2001) small case-study and the more extensive work of Kasesniemi & Rautiainen (ibid.).)

Participants were assured of the confidentiality and anonymity of their responses; this was especially important given the personal nature of the messages. Of the students available, 135 (approximately 70%) of them responded to our request for their messages. The mean age of participants was 19; three-quarters of them were female (n = 120, 75%) and a quarter male (n = 39, 25%). As was typical of the university's intake more generally (with 28% of the students actually from Wales itself), almost all the participants were British (98%). Although Cardiff University attracts a largely White, generally middle-class population of students, there is no apparent reason why the sample used here might not also be fairly representative of young, university-age adults in Britain as a whole. Having said which, even though anecdotal evidence suggests that many other young people their age are equally heavy users of mobile phones and text-messaging, we do not assume that the sample is more widely representative in terms of educational background and socio-economic status.

A total of 544 separate messages were recorded by participants which were transcribed as accurately as possible into a single electronic document. For the most part, we followed straightforward Content Analytic procedures (see Krippendorf, 1980) in organising and interpreting participants' text-messages; this is an approach well-suited to the descriptive analysis of open-ended or qualitative textual data such as ours (Bauer, 2000). For example, in pin-pointing their primary functional orientation, the text-messages were coded in terms of 'referential units' (Krippendorf, 1980:62) - the main, relatively discrete idea expressed in each message. As part of this systematic process of inferential organization, we then clustered all these referential units into broader ideational categories. At no point were the categories necessarily either mutually exclusive or exhaustive; however, we have sought to follow the guidelines of explicitness and 'best fit' (Pidgeon & Henwood, 1997:261) by providing recognisable descriptions of, and examples for, the different categories. Partly given the size of our corpus, we have not felt it necessary to undertake elaborate statistical analyses other than to calculate broad descriptive tendencies in terms of our central interests: (a) message length (i.e. number of words/characters used); (b) main typographical and linguistic content such as emoticons (e.g. :-)), abbreviations and letter homophones (e.g. Gr8 'great', RU 'are you'); and (c) primary functional orientation.