Chapter Three: Methodology and Methods
3.7 Key tenets of Grounded Theory
Regardless of the ontological and epistemological differences, there are some common threads running through all variants of grounded theory (e.g. Glaser and Strauss, 1967; Strauss and Corbin, 1990; 1998; Charmaz, 2006; Corbin and Strauss, 2008). These relate specifically to the management and analysis of data, namely theoretical sampling with concurrent data
collection and analysis, the constant comparative technique, memo writing, theoretical sensitivity and theoretical saturation (Birks and Mills, 2011). It is important to note however that grounded theory is not a process that can be applied with rigid guidelines, rather there needs to be a flexible approach (Stern and Porr, 2011), but it could be argued that without adherence to core processes of the methodology the research could not legitimately be referenced as grounded theory. The following sections will explain how all these elements of constructivist GTM were operationalised in the current study, as well as the methods of data gathering, ethical issues and the cannons of rigour.
3.7.1 Theoretical sampling
The approach to sampling in this study was planned as a two-stage strategy. Sampling would be purposeful in both stages, but in slightly different ways.
The initial sample would be a convenience (but purposive) sample in order to provide some baseline relevant data, that is, participants would be recruited from the research wards who had experience of caring for acutely ill deteriorating patients. Early sampling was not a pilot study and had not been planned as such; rather it was a starting point for analysis from which the early codes and nascent concepts would be needed before theoretical sampling could begin. A potential limitation of this approach is that theoretical sampling might be started too soon, before sufficient baseline data is gathered, with resultant unfocussed or premature categories
(Charmaz, 2014). It was essential therefore to conceptualise relevant ideas before moving on to the next stage. Charmaz (2006:18) warns against
skimpy data in these early stages, so it was important to sample key
informants who could provide rich data at the outset of gathering in order to answer the broad question, what is happening here? Using observational methods of data collection in the early stages of the research proved to be a useful strategy in which the researcher could obtain a broad view of the field upon which later interview questions were posed. Thus, the subsequent theoretical sampling approach would also be purposeful, but one in which the researcher was able to focus on data that helped to fill out developing codes, categories and concepts. The sampling strategy in constructivist grounded theory is therefore directed by the ongoing analysis of the data (Charmaz, 2006) and the researcher role is similar to that of a journalist or detective, following up on leads found in the early data (Stern and Porr, 2011).
Early analysis of the first tranche of data from the initial participants provided some tentative categories that directed the next phase of data gathering. Theoretical sampling then enabled the researcher to pose more focussed questions to the participants and start to sample the data more theoretically. That is, in an iterative process, the researcher had to make strategic decisions about where to look and from whom to gather more data that would develop the categories already identified and further expand on their properties (Birks and Mills, 2011). This functioned as a time saving device, because as Charmaz (2006) suggests, less effort is wasted on aspects of care that have nothing to do with the emerging theory. Table 10 below lists the participant inclusion and exclusion criteria.
Table 10. Inclusion and exclusion criteria for study participants
Inclusion criteria Exclusion criteria
Employed by the Trust at the research site
Working on one of the four designated acute wards (two surgical and two medical)
Over 18 years old Registered nurses Doctors
Physiotherapists
Occupational therapists Pharmacists
Healthcare support workers
Healthcare staff with experience of caring for the acutely ill deteriorating patient
Healthcare professionals not working in acute areas on the designated wards were not invited to participate.
An example from the current study may serve to clarify this process. Some of the early interviewees talked about recognising deteriorating patients because they were not right, so later participants were asked if this was
their experience and if so, what were the early signs that for them constituted being not right. Checking out this principle with different professional groups was a helpful strategy that served to further fill out the category not right. Later on in the data gathering, explaining some of the categories to the participants during interviews to see if they resonated with their experience was a way of ‘member checking’, (Charmaz, 2006:111). Furthermore, Pope and Mays (1995) highlight that where a sample is
theoretically informed and relevant to the research question, this minimises the possible bias arising from selecting a sample on the basis of
convenience. Memos, discussed in section 3.7.4 below, provide an audit trail of decisions made regarding theoretical sampling.
3.7.2 Coding the data
In this section, the initial and focussed coding processes adopted in the study will be explained.
In this study, the researcher typed the transcripts verbatim as soon as possible after the data gathering sessions. Listening to the interview
recordings and typing up observational field notes enabled the researcher to become familiar with and immersed in the data once more. A final check of the completed transcript against the digital record or hand-written
observational field-notes meant that the data was listened to or read at least three times before coding commenced and a high standard of transcription accuracy was established. Arguably however data analysis had already
commenced cognitively during the data gathering as the researcher interviewed and listened to the participant or observed them in practice.
a) Initial Coding
In the first round of coding, data was analysed line-by-line and coded using a word or phrase to capture the essence of what was happening, to label or define the important actions or words in the text, written in the margin on the hard copy of the transcripts or field notes (Saldaña, 2009; Birks and Mills, 2011). Thus verbs were adopted for the codes where possible, as recommended by Charmaz (2006), in order to stick closely to the meanings within the data and capture the action. Researcher-generated codes took the form of gerunds (the noun version of the verb), so for example vigilance
was coded as beingvigilant.
Most codes were generated from the researcher’s vocabulary, labelling what was happening in the data, however codes were also formulated from the language used directly by the participant. Invivo codes thus comprised and preserved the participants’ own words verbatim and their meaning
(Charmaz, 2006:55; Birks and Mills, 2011). Being not right is one example of an in vivo code, a term that several participants used to identify the early stages of deterioration, a condensed, shorthand term used colloquially in the shared vernacular with significant meaning for the participants in this research. These types of codes helped to anchor the analysis firmly in the participants’ world (Charmaz, 2006).
Data was coded for a wide range of activities; practices, episodes, encounters, roles, social types, relationships, groups, behaviours, rules, emotions and hierarchies, but not for themes. Themes, or rather what were referred to as categories in this study, came later from focused coding (Saldaña, 2009). Initial line-by-line coding was quite a slow process, but it enabled the synthesis of large volumes of data into a condensed form. As coding became more focused, patterns began to emerge from the data (Saldaña, 2009) and as Charmaz (2006:57) suggests, the most significant or
frequent codes were identified and grouped together into categories.
Larger chunks of data were then reviewed together, incident-by-incident, and labelled using new codes where appropriate or using the most
significant of the earlier codes. Codes, the single units of analysis, were then clustered together to create categories that synthesized and explained larger segments of the data. Categories, and the way that they linked together, would eventually form the basis of the key concepts used in the construction of the final framework.
Coding and categorising the data was not a linear process, rather it was iterative and cyclical as the researcher moved back and forth between different segments of data, comparing incidents, participants, new data with old data, with line-by-line and focussed coding. With each cycle, the codes and categories became more refined and abstract, providing a more
conceptual grasp of the whole (See Figure 2 below). But most importantly, sticking closely to the data promised to produce a framework that would fit
with the participants’ experiences and have relevance for them in practice (Charmaz, 2006:54).
Figure 2. Coding, categorising and conceptualising the data. This figure illustrates how codes were sorted into categories, and then categories were sorted into core categories and ultimately into key concepts (only two are shown here). Colour coding links to that in Figure 3 presented later.
c) Axial and theoretical coding
Further coding strategies, axial and theoretical, provide a framework that can be applied to data to guide the researcher and add structure to the final analysis. These were not used in the current study and the rationale for this decision is given below.
Axial coding is a process that helps the researcher to specify the properties and dimensions of a category and to articulate how categories are linked or related, pulling together the fractured data (Strauss and Corbin, 1998:124; Corbin and Strauss, 2008). This arguably forces the researcher to apply a predefined scheme, namely conditions, actions and consequences, onto the data. If however, as Charmaz (2006:63) suggests, the researcher can tolerate
Core process key concept core category category codes category codes core category category codes category codes key concept core category category codes category codes core category category codes category codes
ambiguity, the data will provide the direction required and a scheme is superfluous. In the current study, specific factors that either enhanced or inhibited actions within each of the core categories were identified.
Theoretical coding, articulated by Glaser (1978), allows the researcher to select from a list of 18 theoretical coding families. This list was extended in a later publication to include more theoretical codes (Glaser, 1998). Saldaña (2009:163) suggests it is an umbrella term used to cover all codes and categories, the core that helps to integrate the work. Theoretical coding was not applied in this study as it presented a conflict with the underpinning principles of not wishing to force the data into preconceived categories, and as Charmaz (2006) indicates, the data itself will invoke the codes required. She suggests however that theoretical codes can be used if they fit, and they may move the analysis in a more theoretical direction, but one must avoid imposing the framework on the data analysis (Charmaz, 2006:63).
Saldaña’s (2009: 187) ‘touch test’ was useful as a check on the theoretical or conceptual rendering of the data in the current study. He explains that to progress from the real to the abstract, one needs to apply the ‘touch test’. For example, a ‘mother’ can be touched, but the concept of ‘motherhood’ cannot. None of the core categories in the final conceptual framework from the current study can be touched e.g. being vigilant through surveillance,
d) Data management
Glaser (2005) highlights his resistance to the use of computer software for qualitative data analysis and Charmaz (2006) provides no guidance on its use, but more recently Birks and Mills (2011) encourage its use in
conjunction with manual approaches. Initially, manual data analysis was the plan in the current study in order to fully immerse oneself in the data. After coding the first few transcripts however, it soon became apparent that handling a large amount of data in this way would not be feasible. Manual coding continued throughout the study, using paper-printed transcripts, but NVivo7 was employed as a data storage and sorting facility. This data
management package was selected because it was available to students at the University of Warwick at no additional cost. When feeling overwhelmed by voluminous data, a comprehensive visual representation of the whole dataset on a computer screen was a helpful adjunct to the analysis process. The database remains in storage to provide a clear and transparent audit trail of the coding decisions made during the analysis process.
3.7.3 Constant comparative method
The constant comparative technique is a key component of grounded theory and was used repeatedly throughout the analysis process, where the
researcher constantly returned to the data to check developing categories and concepts to guide the gathering of new data where appropriate (Charmaz, 2006). Each new set of data was then compared with the last.
Each incident was compared and contrasted with others in order to focus on the emerging properties of a category and identify patterns within the data. Category was compared with category, incident with incident, participant with participant, groups of participants and data from the same individual were also compared for emerging ideas (Birks and Mills, 2011). Through comparison of the data, either from different or similar groups and events, categories and their properties and their relationships with each other started to develop (Charmaz, 2006). The number of categories was then reduced through the discovery of uniformity between them, giving rise to the core categories, that is, higher-level concepts that provided the building blocks of the framework. Figure 3 below provides an example from the current study to clarify.
Initial coding of the data highlighted codes such as checking, observing, being observant, being vigilant. These codes developed into the category vigilance.
With further comparison different types of vigilance were identified, namely
primary vigilance (carried out by nurse participants), secondary vigilance
(carried out by those with less contact, in a peripatetic role) and layperson vigilance (provided by non-healthcare professionals, such as the patients’ significant others). All of these codes and categories were collectively housed within the core category or concept of being vigilant through surveillance.
Figure 3. An example of codes, categories and core category. This links to Figure 2 above and uses examples from the current study to illustrate how codes became categories, core categories and concept
Core
Process Concepts Core Categories Category Code
3.7.4 Memos
Writing memos has been described as the intermediate step between data collection and writing the draft report (Charmaz, 2006) and in the current study memos served as an integral part of coding the data. For this
researcher, memos were akin to thinking aloud about patterns in the data and, as such, continue to provide a permanent and transparent audit trail of the analysis process undertaken. The researcher read a line or section of a transcript, coded it, and then paused to write a memo about the code (or the category in which the code was framed). All memos were documented with a title, date and referenced to the code or category with which they were associated for further analysis. Thus, thoughts and feelings about the data, codes and categories and how they linked together over time were recorded, as instructed by Charmaz (2006:72:82). This gave the researcher
Making the link
Being vigilant through surveillance
primary vigilance secondary
vigilance layperson vigilance observing by others
Identifying deterioration and recognising urgency
opportunity to think about the data, develop a writing style, discover gaps in the data and generate new ideas for data collection while maintaining an audit trail of the research process, similar in style to a journal (Birks and Mills, 2011).
In a final step, memos were sorted to facilitate the generation of the conceptual framework; by putting all the fractured data back together in such a way that it explained what was happening in clinical practice. The nature of memos, written rapidly and spontaneously in informal, unofficial language for personal use, as advised by Charmaz (2006:80:84), rendered them difficult to write directly into a computer. Pencilled notes were more convenient during the analytic process, where ideas flowed freely and uninhibited, but this presented difficulties during the sorting process. On reflection, typing up handwritten memos into NVivo© would have been time
consuming, but could have eased the sorting process and allowed for the cutting and pasting of key sections of text into the final report.
Clustering and diagramming
Clustering is a technique, recommended by Charmaz (2006:86) to
understand and organise material. As a visual learner the researcher found this technique helped to capture spontaneous thoughts about connections and relationships within the data, sometimes clearer than verbose memos (Birks and Mills, 2011). A central idea or category was written in the centre of a map, with spokes to smaller circles that showed the defining properties and relationships. Figure 4 below highlights a simple example from the
current study. As the study progressed, diagrams became more complex and were used to identify connections between categories and concepts, which ultimately developed into the final framework (Figure 21 in Chapter Four). The process of memo writing and diagramming aims to facilitate the development of theoretical sensitivity.
Figure 4. A cluster diagram. Using an example category from the current study, Something not right, this figure illustrates how diagrams were used to illustrate the code sorting process
3.7.5 Theoretical sensitivity
According to Glaser and Strauss (1967:46) theoretical sensitivity is the ability of the researcher to have theoretical insight into an area of research (and potentially themselves), but also to be able to make something with that insight. This was potentially daunting for a novice in grounded theory, the challenge to think theoretically (Tarozzi, 2011:11), where the threat of being simply descriptive rather than analytical was ever-present. But
Something not right Noticing the difference Looking different from the norm Behaving differently Being with the patient Knowing the patient's norm
Tarozzi (2011) is reassuring in his assertion that producing a theory is a skill that can be taught and therefore learned. Thus, the early interviews in the current study provided the opportune training ground. Personal
reflection upon the increasingly theoretical memos produced, as the study progressed, indicate that learning to theorise is possible. The process of stopping and thinking about the data as it was gathered, comparing it with other pieces of data, discussing with supervisors, making connections and developing new questions to go deeper each time was, as Charmaz (2006) indicates, theorising. This was supported by her advice to avoid coding data for themes because this would result in a simple description of action rather than a theoretical rendering of it. Coding for action throughout from the outset, in the form of gerunds and asking ‘What is this a study of?’ at each stage of the analysis, facilitated the identification of three concepts and a core connecting process (Charmaz, 2006).
3.7.6 Reflexivity
Charmaz (2014) explains that the researcher does not enter the field with