Chapter Four: Methodology
4.5 Analysing the Data
The problem when analysing the data was how to condense highly complex and context-bound information into a format, which tells a story that is convincing to the reader. The study generated many hours of raw data from transcripts (ranging from 10 to 35 pages in length), interviewer notes, observations and documentation. The philosophical background to the study and research objectives were both considered when researching the best method to use to analyse the data. Data analysis techniques reside on a continuum in terms of the level of transformation of data required from descriptive to interpretation (Sandelowski and Barroso, 2003b). For instance, content analysis ‘tests’ theory, as the codes are drawn from theoretical ideas (deductive coding) and therefore considered descriptive. Grounded analysis explores new links as codes are drawn from the data itself (inductive coding), thus providing more detailed interpretation of the data. Themes derived from theory enable the researcher to replicate, extend or refute prior discoveries whilst inductive themes are often useful in new areas of research (Boyatzis, 1998). For the purpose of this study, some a priori
162 codes or themes were drawn from the conceptual framework that comprised the work of Fuller and Unwin (2003); Felstead et al, 2009, but it became clear once the data was transcribed that it would also be useful to work with themes drawn from the data itself. This prompted the search for a flexible method of data analysis that combined both deductive and inductive coding.
4.5.1 Template Analysis
A relatively recent development in organisational and management research is that of template analysis (Crabtree and Miller, 1999; King, 2004; Waring and Wainwright, 2008). It can be used within a range of epistemological positions, including realist qualitative research (King, 2004) although is unsuitable where quantitative and qualitative data is combined, as it may appear to produce coded segments which could simply be treated as units of analysis for content analysis. The attaching of codes to segments could also prove limiting to discourse analysts who want to explore the meaning and ambiguity in the use of language. A more detailed description of template analysis follows below.
Template analysis refers to a varied but related group of techniques for thematically analysing qualitative data (King, 2004). It has emerged from more structured approaches such as grounded theory and phenomenological analysis (Waring and Wainwright, 2008) and may be used to analyse any form of textual data including diary entries, interview transcripts, electronic text etc. However, it is less prescriptive than grounded theory, in that it does not specify procedures for data gathering and analysis (Strauss and Corbin, 1990). It provides a more flexible technique with fewer specified procedures, permitting researchers to tailor it to their requirements. King
163 (2004) suggests this type of analysis works particularly well when the aim is to compare the perspective of different groups of workers within a particular context (King, 2004).
When using template analysis the first stage was to develop a coding ‘template’. The epistemological position of the researcher will often influence the selection of codes. A ‘half way’ position was chosen to select the codes (Miles and Huberman, 1994; King, 2004). This involved starting with some a priori codes taken from the interview template and strongly expected to be relevant to the analysis. In this instance taken from the literature review and my experience of teaching on the FdA in Retailing, these codes were then refined and modified during the analysis process (Crabtree and Miller, 1999:167).
The next step involved transcribing the interview recordings and other primary data taken from observation, researcher notes and documentation. Voice recognition software was used when transcribing interview recordings. Transcribing the interviews verbatim was important as it facilitated familiarisation with the data prior to starting the coding process. Thereafter each transcript, field note and observation record was read line-by-line in order to identify potential themes and thoughts, comments were attached to each. This proved very time consuming but was a necessary and valuable part of the process. The next step comprised initial coding of the data. Realism research is not interested in every detail of all the perceptions of respondents, like constructivism research, instead they are only interested in those perceptions relevant to the external reality (Sobh and Perry, 2005:1204), only relevant parts of the transcripts were therefore highlighted. This served to pull together and
164 categorize a series of otherwise discrete statements, events and observations. NVivo 9 software was used in the early stages to aid the process of data analysis, selected over other packages as it works well with large data sets in retrieving and coding data (Jones, 2007). Guidelines for using computer aided data systems were followed, this involved gaining a good understanding of how the system operated by attending NVivo software demonstrations (Blismas and Dainty, 2003; Welsh, 2002). The use of technology for data analysis purposes has received criticism as some fear it will take over the analytical process when it should be the researcher’s responsibility to extract key codes and concepts (Mason, 2002), furthermore, computer software may reduce sensitivity to important aspects of realism research that emphasize relationships, connections and creativity (Carson and Coviello, 1996). NVivo 9 provided a useful depository for the interview transcripts and proved helpful in the early stages of data analysis. However, problems using this software started to emerge when writing-up the findings chapters and at this point it was unclear how best to communicate and present the wide range of data collected. It took three iterations to organise these chapters, prompting the search for a new data analysis strategy that acknowledged the existing literature while simultaneously remaining open to possible new findings, what Strauss and Corbin (1990) refer to as a mix between analytic and emergent categories. From this point, the data was analysed manually.
When identifying codes manually it was necessary to record the relevant code, in the margin of the transcript using different coloured highlighter pens. An A4 notebook proved useful in recording where codes could be found across the data set. Participant’s names were listed across the top of the page so that when codes were later identified in relation to each participant they could be added to the column
165 below. The recorded information comprised page numbers, from transcripts, observation notes and documentation, and a phrase or two for clarification. If a section of data encompassed one of the a priori themes then it was appropriate to attach a code. In instances where no relevant theme existed, a new one was added or an existing code was modified (See Appendix D, for the template used on the theme of support). Changing the template to accommodate the text made analysis more inductive. As the process continued more themes emerged, this resulted in going back to previously coded data to see if there was any evidence of information previously not recognised as significant, this avoided the risk of missing information about the underlying structures and mechanisms. The coding exercise was initially undertaken independently, however later modified as different iterations of the findings chapter emerged during the write up stage.