3. THE CASE
3.5 APPROACHES DRIVING THE ANALYTIC PROCESS: DATA ANALYSIS
3.5.3 CODING
There are phases of coding that have been described differently over grounded theories’ history, so I look to a recent review and summary of these from Birks and Mills (2011). They start with initial coding, moving to intermediate and culminating in advanced coding, where the movement between them is directed by theoretical sampling (Birks and Mills, 2011). I now review the coding phases in relation to my analysis of the data.
In the initial phase of analysis I use line by line coding of each transcript, to fracture the data (Birks and Mills, 2011; Charmaz, 2006) to create separate pieces of the text (Bazeley and Jackson, 2013), in order to see it critically (Charmaz, 2006). Coding data as gerunds, noun forms of verbs, using words that end in ‘ing’ can help to prevent making conceptual leaps, but also develop concepts and an abstract level of analysis, moving away from purely descriptive codes (Birks and Mills, 2011; Corbin and Strauss, 2008; Charmaz, 2006). For example I used either single words or short phrases in my coding of the transcripts. So, in the case of Tess I used codes such as developing, challenging behaviours, learning, moving, conflicting reports. For Liz I used developing practice, team thinking and sourcing information. Gladys’ codes included engaging with occupational therapist, rapport building and collaborating. Each line that I coded using NVivo10 was linked to its original transcript and could be traced back to the transcript at any point.
I sometimes used in vivo codes derived from the practitioners’ words from the interviews to represent a code or category (Charmaz, 2006) where I could not easily think of a gerund or could not capture the detail easily and where the practitioners’ words were clearer. For example: “almost being suckered into feeling” (Tess IV4) and “occupational therapy is not fluffy” (Liz IV6). This coding moves my analysis to the next phase, explained next.
Intermediate coding is about sorting, organising and synthesising the codes already created (Birks and Mills, 2011; Charmaz, 2006). NVivo10 allowed me to look at line by line codes of interest in relation to the longer passages to which they belonged (Saldana, 2009), which provided me with more context about what I had coded. I organised my codes by initially keeping one folder of codes per one transcript per practitioner. I then combined all the nodes from all the transcripts for each practitioner into one folder. Once all transcripts had been coded all practitioners’ codes were combined in one folder. This method of storing the codes allowed me to analyse the data within any one transcript, across the codes of two or more transcripts per practitioner and between one or more transcripts across all practitioners.
There is however, a problem with using multiple folders. For example I have one folder for Liz’s interview three that I coded and then added to a folder of all of Liz’s codes from all twelve interviews. I would finish with all of Liz’s codes, from all of her interviews being placed in a folder along with all of Tess and Gladys’s codes. The codes are repeatedly copied into each new folder and this can give a false impression of what and how much data is coded and available for analysis (Bazeley
and Jackson, 2013). Ultimately I did not use the analysis tools, such as modelling
for visualising the data to make connections and identify relationships (Bazeley and Jackson, 2013). The small number of practitioners and the ease with which the codes and folders can be organised, with regular memos, meant that I was able to manually develop concepts and categories along with figures created in Microsoft Word (see appendices 9-11a and b). This matter is discussed next.
I develop concepts and categories from the codes. “Concepts are words that stand for ideas contained in the data” (Corbin and Strauss, 2008, p 159) also they are interpretations and the results of analysis (Corbin and Strauss, 2008). I collated the related concepts with each of the developing categories, in order to refine them and to facilitate further analysis to establish how they were related. Concepts are composed of both properties and dimensions. The former are the “characteristics that define and describe concepts” (Corbin and Strauss, 2008, p 159) and the latter are “variations within properties that give specificity and range to concepts” (Corbin and Strauss, 2008, p 159). I initially used the NVivo10 sets tool to create categories that contain and hold the concepts and codes supporting them, for example some
early categories were core skills, environment, risk, challenging behaviour, client- centred practice.
There were three practitioners with between seven and twelve interviews per practitioner, and sets can help manage this amount of data (Bazeley and Jackson, 2013). Sets provide shortcuts to any code and can document and hold the items together without merging their content, thus data may also belong to more than one set (Bazeley and Jackson, 2013). As I create each concept with its properties and dimensions and each category, I make a memo to summarise each one (appendix 14) that helped to reduce the concepts and combine them into an emerging abstract theory (Charmaz, 2006). I wanted to see all the codes from all practitioners combined, having completed the sets and related memos about each category. I therefore went back to the combined codes folder and created a code that represented each category and then moved the codes for the associated concepts to their respective category code. This enabled me to see all categories and related concepts, based on the coded data in one place.
I organised the codes into concepts and then into categories according to shared characteristics and patterns in the data (Saldana, 2009). I included a range from Saldana (2009) that included similarity, frequency, causation and correspondence, considered next. Similarity was where things happened in the same way, and emerged in codes about how the practitioners performed their assessments and in the difficulties surrounding their use of the required standardised assessment. Differences are concerned with how actions can happen in predictably different ways. I found the practitioners’ variation in sequences and certain orders of actions and decisions were revealed on the timeline. I subsequently found that the occupational therapy process was not followed in a linear fashion. I considered the frequency of how often or seldom actions occurred and my analysis indicated that the practitioners tried to use a specific assessment to measure the service users’ occupational participation. Not only that, but the time when assessments were done and how they were enacted showed how seldom the practitioner’s used the assessment. As a result of the limitations of the assessment, the practitioners used their observation skills in order to assess the service users. I also found Saldana’s (2009) characteristic of causation was apparent from when the practitioners used the standardised tool and their experience of the limitations of it appeared to cause other actions such as finding an alternative, so they used observation. I reviewed
how each practitioner practised in their respective settings such as when using core skills and risk assessment and management. Consequently I found they indicated Saldana’s (2009) characteristic of correspondence with other activities or events.
The advanced level is the grounded theory at its most abstract and generalizable (Charmaz, 2006). My concepts are initially formed using terms associated with occupational therapists’ practice of such as assessment, risk management and intervention. Thus I tried to make the specific dimensions and properties of the concepts within categories increasingly explicit (Birks and Mills, 2011). As the analysis develops I make figures (discussed later) that include the terms mentioned looking for relationships between sub-categories and categories (Charmaz, 2006). I develop more abstract categories, including frameworks, processes, personal and professional (see appendices 16-17) for conceptual depth and breadth, in order to give a more theoretical view of the data (Birks and Mills, 2011). I develop the final
categories to be more abstract and they are presented in the findings. In order to
develop and refine categories the data generation has to become increasingly focused, which is achieved using theoretical sampling, discussed next.