3. Methodology and theoretical framework
3.3 Methods for conducting a Frame analysis
3.3.1 Frame validation
The method of identifying and analysing frames for this thesis is informed particularly by the descriptions of two scholars who have conducted frame analyses in very different settings and on very different data. The first of these is Strong (1979), who examined the ‘ceremonial order’ of the medical clinic; the second is Koenig (2006) who applied a frame analysis to transnational public discourse. Both provided explicit discussions about how to deal with a central problem in qualitative analysis in general and frame analysis in particular; that is, “How can we be certain that such and such really did mean what I said it meant?” (Strong, 1979: 230).
Strong’s question is at the heart of questions of validity in qualitative research which have been debated by others (Whittemore, 2001). While quantitative research can rely on a realist approach that defines validity as the closest possible approximation to reality, a constructivist approach emphasises, instead, understanding and elucidation, recognising that this may be only one of a number of valid interpretations (Cohen & Crabtree, 2008). In taking a
constructivist approach, I favour the position proposed by Eakin & Mykhalovskiy that validity in qualitative research should be considered less in terms of procedure and practice, and more in terms of how the analysis is “rendered convincing”, drawing on appropriate data and making claims for which evidence can be reasonably claimed (Eakin & Mykhalovskiy, 2003).
Strong answered his own question by pointing out two fundamental premises for using any kind of discourse analysis as an approach to studying social phenomena. First, that because the social world relies on shared access to resources for interpreting actions, these resources should also be accessible to the researcher, and to readers of that research. The same principle guides Silverman’s approach to his analysis of ‘social relations in the clinic’:
“Inevitably, I must concede that, in highlighting situational factors, I have drawn tacitly on everyday knowledge about forms of talk and what they mean. Thus, if readers are able to detect the patterns in talk which I depict, this will be because we are using the same commonsense resources.” (Silverman, 1987: 21)
Second, that through detailed examination of several instances of the phenomenon being studied - contrasting, quantifying, looking for differences and similarities, seeking to explain deviations – it is possible for the analyst to make explicit those bases of action and
interpretation which govern interaction but which are rarely explicated by participants themselves.
Strong describes the process by which he generates and tests hypotheses concerning the rules of interaction which he observed. He uses analytic induction to generate his theory from the data itself, rather than from existing literature or a pre-formed system of codes:
“Explanations are tested out all the time, trying them on one case to see if they fit, then moving on to another, and so on. Where the proposition does not cover the particular case, it may be reformulated or else the phenomenon redefined until a universal relationship is established which fits all the cases.” (Strong, 1979: 234)
The process for this testing is a continuous questioning and re-evaluation of the codes being developed from the data. The categories developed are applied to each part of the data to see if they fit. Where they do not fit, categories are reformulated or added, to incorporate the new data. In this way, the ideas brought by the analyst are tested against the data and changed. The process is reiterated until a consistent set of codes is found which adequately fits the data.
With the large dataset at Strong’s disposal – a total of 1,120 consultations were observed – he was able to conduct this process on one half of the data, using this to formulate an argument as to his findings. The other half was then analysed to test and refine this argument.
Koenig (2006) attempts to address a perceived lack of rigour in applying frame analysis by proposing a four step model. He recognises the criticism of frame analysis that new frames are often identified – “or even fabricated” (Koenig, 2006: 62) - by each researcher with little attempt to establish validity of these frames.
Koenig recommends that initially a sample of the data is analyzed using a discourse analytic approach until no new frames and themes are identified. Koenig’s focus is on constructions of identity through the text. This may involve detailed analysis of language used, for example the usage of first person plural pronouns (Koenig, 2006: 65). Identification of frames needs always to address the broader theoretical aims of the research. Step two involves testing and
developing the emerging frame model on larger datasets. Koenig proposes using software to facilitate this process across a very large dataset. From this stage, a reliable frame model is developed which can then be applied to the entire dataset systematically in step three. For Koenig, this stage relies on the identification of keywords that represent frames. While acknowledging the difficulty of identifying and validating these keywords, Koenig depends on these to explore how frames are distributed across a very large dataset. Finally, Koenig
proposes a form of statistical validation based on the occurrence of these keywords. The use of keywords in this sense is helpful where large datasets is involved, and may produce a useful numerical overview, but the validity relies heavily on being able to identify useful terms and to reduce concepts to single words or short phrases. I would question the ability to do this with many datasets, and certainly I do not feel that this would be possible with data like that which is drawn on in this thesis.
Both Strong and Koenig are attempting to develop and describe a systematic process for identifying and validating a structure of frames. While Koenig takes a statistical and in many ways reductionist approach, Strong’s analysis is more qualitative and exploratory. However, both highlight some key principles for conducting a frame analysis:
1. This form of analysis is not about imposing a predetermined set of frames upon a dataset, or fitting the data into a preformed theory. Frame analysis works best as a form of analytic induction, informed by the participants’ actions and the significance they apparently attach to them;
2. Frames do not take an identical form in every instance. As Strong points out, “the same rule may be embodied in a great variety of very different behaviour according to circumstance” (Strong, 1979: 232). As such, the more data that is analysed, the more these rules and categories can be refined and tested, so that they begin to approach generalisability.
3. For findings to be credible, the process of analysis should be made as visible as possible.