Product Information:-

  • Journals
  • Books
  • Case Studies
  • Regional information
Request a service from our experts.
Visit the JDAL journal page.

How to... implement grounded theory

Options:     Print Version - How to... implement grounded theory, part 3 Print view

How to proceed with GT

As seen from the brief summary of the two schools in Table II, there are subtle variations in both orientation and procedure, and anyone using GT seriously should immerse themselves in the literature about the approaches for a more detailed account than is possible in a short article. Jones and Noble’s 2007 critique of GT’s use in management research provides a good introduction.

The key difference between any sort of GT and other types of management research is the speed with which the researcher immerses him or herself in the data. The initial stages of selecting the question and designing the research will inevitably be both shorter and iterative, as other questions and other approaches are thrown up during the course of the research.

However, Partington (2002: pp. 138-143) advocates that before data collection the researcher needs to look at four fundamental issues:

  1. What is the general purpose of the research? For example, are you attempting to explore, describe, understand, explain or change a particular phenomenon?
  2. What is/are the research questions?
  3. What is your theoretical perspective?
  4. What is the design of the research? Whereas for other types of research, the design is clearly thought out at the beginning, that for GT tends to be "sketchy and incomplete" (Partington, 2002: p. 143) because of the use of theoretical sampling (see above) and the way the data collection strategy is driven by ideas.

Another distinctive feature of GT is the way in which the data analysed can come from a variety of sources, quantitative as well as qualitative: interviews, observation, documents, or even surveys (Geiger and Turley, 2003).

Daengbuppha et al. (2006) used mainly in-depth ethnographic interview, participant and non-participant observation for their study of heritage sites, but they also used published material and archive data.

Pauleen et al. (2001) used semi-structured interviews and discussions with the participants, the researcher’s journal, participants' notes, organizational documentation, and e-mails, for an action research study.

Macri et al. (2002) conducted a grounded study of change in a small organization, with participant observation as their main method, but they also used interviews and documents relating to the region’s economy, the organization’s competitors, internal documents such as memos, calls for meetings, meeting excerpts, selling and delivery records, and consultants’ reports, as well as the company balance sheet.

Data collection, coding and analysis

This is clearly the heart of GT, and should be started as soon as possible. The processes of collection, coding and analysis are iterative and so as soon as the researcher has data, he or she should start coding and analysis.

If using interviews, the questioning technique should be as open as possible, to allow for the participant’s experience to emerge undiluted.

If using case studies, the number should be kept small given the practicalities of applying such an in-depth approach.

Partington (2002) recommends recording all interviews and transcribing them oneself, to ensure maximum familiarity with the data, as well as to allow the opportunity for reflection on interviewing style. Coding software can be used, although many researchers prefer manual analytical techniques such as cards, sticky notes, etc., and having to learn a new software package can be an added distraction for a researcher new to GT.

The first task of coding is to break the data into small pieces (analysis is usually line by line), noting the concepts that emerge, which are then provisionally sorted into categories. At this stage, open coding is normally followed: the categories should emerge purely from the data, and the researcher should keep an open mind.

Wastell (2001) conducted 26 interviews with the questions "Tell me what leadership means to you", and "How do you feel about your job?". The initial four interviews generated 51 concept cards which were sorted into nine categories. The main category was "subordinate conduct" with two subcategories, "leader influence" and "subordinate attributes".

Douglas (2006) conducted research into managerial decision making and found 15 categories with a total of 200 properties. From this, two central constructs evolved: "management decisions and consequences – employees' perspectives" and "self as manager – manager’s perspective". The former were conceptually categorized as “growth decisions”, “process decisions” and “reaction decisions”.

Categories can be further divided into distinguishing properties:

In Wastell’s (2001) example, the properties of leader influence were: leader contact, information sharing, relationship work, and taking advantage; those of subordinate attributes were: dealing with incidents, making excuses, opinion sharing and taking advantage.

Dimensions are the spectrum within which a property operates, for example leader contact could have the dimension frequent/infrequent.

The process of constant comparison, one of the uncontested principles of GT, occurs throughout.

Sternquist and Zhengyi (2006) describe how they first categorized the first interview, and then read the following interviews to assess whether or not the coding fitted the categories. If it didn't, then either new categories were added or they broadened the existing category. This process was continued until no new categories emerged.

Categories are also compared with one another in order to determine relationships and possible groupings. This is known as axial coding, a process "whereby the provisional categories [are] examined and compared with each other to identify any natural groupings that existed" (Goddard, 2004: p. 549).

Bakir and Bakir (2006) describe how they conducted axial coding in their study of strategy in leisure organizations.

We analysed each category, using the axial coding procedure of the “coding paradigm” (Strauss, 1994, pp. 27-8), where concepts that relate to a category are classified as that category's properties, context, causal conditions, intervening conditions, actions/interactions or outcomes/consequences. This resulted in cumulative knowledge about relationships within the category and between categories.

Selective coding is the relationship between categories and the core category: what Geiger and Turley (2003) call the "storyline".

Geiger and Turley (2003) conducted research on client knowledge of salespeople. They used in-depth interviews and observations of sales calls and team meetings. Initially they coded with a completely open mind, and far more emerged about relationships than knowledge. The next stage was axial coding, when researchers focused on "discovering higher order connections between categories". The key concept was discovered to be "relationship type".

At the selective coding stage, they looked for the data’s storyline and any gaps. These gaps prompted further fieldwork. For example, they wanted to explore relational differences between service and product industries, and the impact of relationship type (for example sectors where the hard sell is the norm).

Saturation occurs once all categories can be subsumed into a core category, and all data fit the emerging theory. The core category is described by Seldén (2005: p. 118) as "a nucleus, a core category, a model, or a central category for others to circle around".

It is very important to avoid closing the research too early and to look for instances which don’t fit the theory.

In Wastell’s (2001) example, the researchers realized after the first four interviews that they were not focusing sufficiently on behavioural issues, so that further questioning and subsequent coding reflected the perspective that subordinates wanted to be liberated from organizational constraints and be more creative. A further category of "facilitative leadership behaviour" was developed to capture the concept that subordinates can feel "limited", "overloaded" or "unleashed". By the 16th interview they had reached saturation point on "‘limited" and "unleashed" but required more work on "overloaded".

Analysis is not, however, done purely by data coding; memos are important tools used to record the researcher’s thoughts about the data as they occur, while diagrams can be a useful way of showing relationships between concepts. Both should be dated and coded in some way so as to link in with the data. They form an important part of the final theory building and the process of integrating the categories into an overall theory.

The essential task of writing the theory is to make the research coherent and comprehensible to others, giving the explanation a structure (Mansourian, 2006). There should be some sort of "audit trail" so the reader knows how conclusions were arrived at.

The following example, quoted directly from Goddard’s (2004) account of his research on local government accountancy practices, provides a useful and detailed case study of data analysis in practice, drawing on the Strauss and Corbin tradition.

"Using Strauss and Corbin's (1990) approach, data was analysed through various stages of coding to produce an ordered data set which was integrated into a theory. Open coding of interview, document and observational data commenced with the identification of provisional categories, which are, “the early conceptual names assigned to data fragments” (Locke, 2001). These categories were identified partly from a line by line analysis of interviews and notes of meeting attendance. A paragraph level analysis and inspection was also undertaken of documentary evidence such as minutes of meetings, organizational reports and manuals and other organizational publications. The emerging categories were compared between interviewees and other sources within each case study to ensure internal consistency. A separate set of categories was initially developed for each case study. A comparison was then undertaken between case studies to provide a set of broader, yet grounded, categories.

The process of axial coding followed, whereby the provisional categories were examined and compared with each other to identify any natural groupings that existed. The groupings which were derived solely from immersion in the data and captured substantive aspects of the research situation are termed substantive categories. They represented another level of conceptual generality in the data and were allocated appropriate labels. Other groupings were derived from the researchers' own disciplinary sensibilities that introduced sociological and organizational meaning to the data. These are termed theoretical categories … Extensive use was made of theoretical memos and diagrams to identify theoretical implications of categories and of the relationships between them. Such memos were also used to develop the final grounded theory. It should be stressed that the above analyses were undertaken iteratively rather than merely sequentially. The validity of categories and groupings was achieved partly by the process of theoretical sampling whereby data was collected from a variety of sources as outlined above and compared within and between cases, and partly by the process whereby emergent findings were fed back to new interviewees for validation, until theoretical saturation was achieved. This was the point where new sources of data no longer provided new information either in terms of identifying new categories or refining existing categories or in the relationship between categories.

The final procedure was the process of selective coding. This requires the selection of the focal core category, that is, the central phenomenon which has emerged from the axial coding process. All other categories derived from the axial coding process must be related in some way to these focal core codes, either directly or indirectly." (Goddard, 2004: p. 549).

Goddard goes on to explain how the selective coding process was achieved by the use of Strauss and Corbin’s "paradigm model", which is a way of analysing the core concept.