As discussed above, coding of transcripts took place using Qualitative Content Analysis (principally following the methods of Schreier (2012)). NVivo v.10, the standard tool recommended and used by the University, was used to facilitate this coding process.
The unit of analysis was chosen to be one interview and the context unit of an answer/question block. No obvious regular unit of segmentation was seen (since sentence structure was coder-imposed during the transcription process and multiple themes could be introduced within one short block), thus codes were assigned to phrases or blocks of text and these blocks separated in NVivo using carriage returns. Each block was assigned a code, which was either substantive (true data) or procedural (question, “social” conversation, clarification or off-topic).
Each interview’s data was added to the database immediately after transcription, however one interview at a time was coded in its entirety prior to the next unit being coded. It was not possible nor particularly intended to directly link categories to specific interview questions (see Schreier 2012, p.79), as it was found that (catalysed by the semi-structured format) interviewees
spontaneously introduced and interweaved many of the topics selected for discussion. Seed categories were therefore created that followed particular research question themes, however these were purely used as an entry and were regularly re-visited, expanded or condensed as the analysis continued.
As data was coded, a commentary “memo” track was maintained in NVivo separately from the textual sources where comments, practical notes (concerning mainly creation, deletion and combination of codes) and concepts identified were stored. From this it was identified that a new type of interview (educator) would be helpful. During coding, additional codes were created only when this could be justified to avoid an unmanageable frame, and rationalised where codes appeared to be both too infrequently used to be justified and their low use was not in itself considered to be particularly instructive. Clearly, as new types of interview were added new codes tended to be generated more frequently, both because the questions were specific to the type but also because the different contexts often brought changes in perspective and priority. Codes were gathered into major and minor categories in a three-level hierarchy, which was a highly iterative process since in reality some early categorisations were unsuccessful in establishing unique dimensions of thought and needed to be adjusted.
As with all such endeavours the frame represented the “best fit” distribution of general concepts across the major organising labels. A very widely-applied code for example was “Professional means...”, which was originally intended as a placeholder for problematic sections which proved difficult to code atomically. The attempt to break this data down into smaller codes however was somewhat unsuccessful. Due to the success of the “list of traits” model of professional status, most answers simply listed these traits, or very similar but nuanced variations on similar themes, to the point that individually coding traits would have left the text scattered as blocks of a few words over potentially dozens of individual codes. As the intent of the research is to look at the effect of the professional model and its homogeneity or otherwise across actors, this would have been unhelpful as there would have been few ways to truly see the notions of professional status side-by-side for comparison. Instead, the widely-applied code was left, then memos taken from an analysis of the text so coded, forming the basis of the following analysis.
Following the coding of all interviews, a further pass was made of all the data (the frame by this point having been established and stabilised) to ensure that data which had been coded in some cases over a year apart were reasonably consistent and to identify whether any further rationalisation could usefully be done. Further memos were added during this process. The first cycle analysis was then undertaken, where for each sub-category the data assigned to that category was read through and summarised, and combined with the memos relating to that data
to form an initial analysis organised by major frame category and then sub-category. The data for each category was biased towards the codes which had been seen (mainly by coding frequency but also by perceived theoretical significance) to be the most significant for the discussion. During this process additional memos were created and a working narrative constructed from the more abstracted notes taken in the earlier stages. The first stage analysis is reported in the next chapter.
A second-stage analysis was then made using the narrative and memos created during the previous work, whereby the major novel and unexpected contributions to knowledge were identified and grouped according to some overall themes. This allowed the tying together of elements from amongst the various branches of the frame into unified areas of theory, which were then argued and summarised in conclusions. This forms the basis for the secondary analysis in Chapter 6. The principal conclusions of the secondary analysis are then summarised and re-stated in the final chapter in juxtaposition to the original research questions.