Chapter 3 Methodology
3.10 The interview phase
3.10.5 Interview data analysis
3.10.5.1 Coding and analysing the interview transcripts
There are a number of different analytical approaches to coding interview transcripts such as grounded theory (Glaser and Strauss, 1967), interpretative phenomenological analysis (Smith et al., 2009) and thematic analysis (Braun and Clarke, 2006). Interpretive phenomenological analysis was considered to be
inappropriate for the context of this study given its focus on language and
linguistics. Rather than strictly following one analytical method, I selected elements of both grounded theory (Glaser and Strauss, 1967) and thematic analysis (Braun and Clarke, 2006) with the aim of developing a systematic approach to coding the transcripts. I also adapted a coding strategy developed by (Saldaña, 2013) who recommends dividing coding into specific cycles using an iterative process. The computer-assisted qualitative data analysis software (CAQDAS) - NVivo was used throughout the coding process as well as manual coding techniques such as writing notes on the text I was analysing and using highlighter pens to indicate possible patterns. Two major coding cycles were used to develop the final themes and conceptual categories from which to base the write-up of the study.
3.10.5.2 First cycle coding
The first level of analysis began by breaking down the data into units of analysis or segments of text. The segments included a sentence, a paragraph or a single word, however, the unit of analysis was determined by the content of the text rather than the length of the text. For example, if several sentences within a paragraph
contained more than one code, the unit of analysis was a sentence or in some cases a single word. Each segment of text was then given a code name. This approach is similar to open coding in ground theory and is commonly used in qualitative research (Attride-Stirling, 2001; Braun and Clarke, 2006; Miles et al., 2014). Further, as Braun and Clarke (2013) point out, coding is not an exclusive process where extracts of text can only be coded in one way. Therefore, where it was deemed appropriate, two or more different codes were also applied
simultaneously to the same data segment to enable ‘multiple meanings’ to be considered in the analysis (Miles et al., 2014). The following extract from Bob’s interview provides an example of first cycle coding and the associated units of analysis (Table 3.2).
Table 3.2 Units of analysis and allocated code name
My earlier thinking about coding initially led me to preference data driven codes over researcher derived codes. This is because the research relating to the education of children with CF is limited and any predetermined codes that are based on literature relating to the education of children with medical conditions generally, could give rise to an a priori coding system that is ill moulded to the data (Miles et al., 2014). I was also reluctant to develop a list of researcher-generated codes based on my own predictions of what might be found in the data as I was concerned that I may not have the objectivity needed for this task given my own experiences of living with CF. However, following an early attempt at coding some of the interview transcripts I began to notice some thematic similarities with the existing literature on the education of children with medical conditions. Therefore, these themes provided an additional focus during the first cycle of coding.
3.10.5.3 Transition to second cycle coding
Following the first cycle of coding, the next analytical activity employed an
organisational approach called ‘code mapping’ (Saldaña, 2013). Two iterations of code mapping took place before proceeding to second cycle coding. As Saldana (2013) suggests, one advantage of having several iterations of code mapping is that it provides an auditing process. Consequently, I was able to document how a list of codes became categorised, re-categorised and conceptualised throughout the analysis (Saldaña, 2013). The first iteration of code mapping involved the
Unit of analysis Code name
‘Yeah, I’ve always felt that people don’t really understand’ ‘Understanding CF’
‘Definitely morning and evening because if I ever had to have anything during the day I would always have it as soon as I got back from school so. I live quite close to the school I could just walk up home at about 3pm and have the medication I needed at night a little bit later’
‘Doing treatments away from school’
‘I didn’t like to get changed in front of all the guys because I
always felt like I was too thin’ ‘Body image’
‘Yeah, my form teacher was the one that really pushed for me’ ‘Form tutor supportive’ ‘Eventually you can’t lie forever, you have to tell your mates,
creation of a simple list of all the codes from the first cycle of coding. This yielded a total of 159 codes. During the second iteration, the 159 codes were compared and sorted to look for replications and to determine which codes seemed to group together. At this point some of the codes were discarded and others went on to form main themes. A visual representation of this procedure is presented in Figure 3.3.
Figure 3.3 Example of second iteration code mapping
This process yielded a total of 55 codes that required further organisation and analysis during the second cycle of coding.
3.10.5.4 Second cycle coding
Saldana (2013) suggests that the goal of second cycle coding should be to
transform existing codes into categories and subcategories which will then progress towards major themes and conceptual categories and then into assertions arising from the research. This is similar to a stage of thematic analysis which focuses on sorting all codes into potential themes and collating all the relevant units of data into the identified themes (Braun and Clarke, 2006). To this end, I began to consider how the existing codes might be combined to form overarching conceptual
categories. This involved looking for similarly coded data, grouping the codes under a theme and attributing meaning to the group by giving it a name. As the second coding cycle continued and codes and themes were reviewed at the level of the entire data set, I found that some segments of text did not entirely fit the code that had been assigned. Where this was the case, the text was assigned to a code that was a better fit or codes were reworked, renamed or removed. At the end of this process, the themes that captured something important about the data and were
Being like everybody else I always want to fit in Don't make them different Being like everybody else Normal Keeping CF private Hiding CF Not telling others Avoiding medication
related in meaning to other themes, were grouped together to form conceptual categories. In some circumstances a sub-category was devised. The following diagram exemplifies the second cycle coding process (Figure 3.4).
Figure 3.4 Example of second cycle coding process
This process of refining and reviewing codes and themes was followed until I developed a satisfactory thematic map of all the data (Braun and Clarke, 2006). This yielded a total of 7 overarching conceptual categories and 23 themes. Clear and operational definitions of all the conceptual categories were established that were specific enough to be discrete and broad enough to encapsulate a set of ideas contained in numerous units of analysis (Attride-Stirling, 2001). The categories and associated definitions were subsequently recorded in a codebook (see appendix 10).
3.10.5.5 Establishing trustworthiness of the interview data analysis
The use of qualitative approaches in research acknowledges that there are multiple realities, and therefore, reliability is argued not to be an appropriate criterion for judging qualitative procedures such as coding (Braun and Clarke, 2013). That being said, there are procedures that can be followed in order to establish the
trustworthiness or ‘dependability’ of the data analysis methods (Braun and Clarke, 2013). One such approach is to engage another researcher in the coding activities to confirm the robustness of the analysis (Strauss and Corbin, 1998; Braun and Clarke, 2013). In this study, three colleagues with experience of conducting
qualitative research cross-checked the conceptual categories derived from my own analysis. They were provided with a selection of the interview transcripts and asked to assign three specific conceptual categories to the data contained within the codebook. This process allowed all conceptual categories to be independently checked across a sample of the data set. Discussions with colleagues following the checks resulted in a change to one of the conceptual categories. It was agreed that
that the category; ‘Support for Learning’ should be renamed ‘Educational Support’. This was because the issue of ‘learning support’ may have had negative
connotations for the participants, and it was not explicitly discussed in the dataset. All other categories assigned to the data were agreed upon by colleagues who checked my analysis.