As the overall purpose of this thesis is to investigate a social science phenomenon with insufficient knowledge, this analysis will draw on the grounded theory research strategy, as it provides a systematic way of working with large amounts of empirical data (Punch, 2013). The data analysis section is designed as follows: Firstly, I will give a brief introduction to the concept of grounded theory and its origins. This is followed by a description of the different steps I have taken in the
analytical process. Throughout I will provide case examples that show my methodology in practice in order to increase the transparency of how the analysis led to the specific findings.
4.3.1 Theoretical underpinnings of Grounded Theory
Grounded theory (GT) is a research methodology that seeks to generate new theories based on empirically collected data in an inductive way (Punch, 2013). With The Discovery of Grounded theory, Glaser and Strauss (1967) sought to answer the critical question of “how did you conduct your research” by demonstrating their rigorous methodological approach, in order to strengthen the validity of qualitative studies vis-à-vis quantitative studies (Punch, 2013). While one school of grounded theorists sees GT as a way of discovering new things from data that is separate from the social scientist, social constructivists see GT as a theoretical interpretation of the data and thus acknowledges the role of the researcher (Charmaz, 2006). Following the philosophic scientific approach detailed earlier in this chapter, this thesis will build on the latter understanding.
The GT research approach centres around an iterative coding process, which aims at categorizing segments of data with a short label to accounts for each piece of data, thus laying the foundation for the analytical work (Charmaz, 2006). Here the pragmatic approach becomes evident, as the most useful data in relation to the research question constantly was identified and used in the further research steps. Although different scholars use slight variations, the overall steps in GT are:
Open coding, Focused coding, Axial Coding and Selective coding (Charmaz, 2006; Gioia et al., 2013;
Langley & Abdallah, 2011). A key element in GT is the constant comparison of data with data, of data with categories and codes, and looking at the interconnection and relation between emerging categories (Charmaz, 2006). Throughout the coding the NVivo software (NVIVO, n.d.) was used to assist my coding process to organize the data and to write memos and to move between the iterative steps. Memo-writing can be seen as the step between data collection and data analysis and a useful tool in moving towards the final framework and can thus be seen as a record of the research and analytical process (Charmaz, 2006).
4.3.2 Open line-by-line coding
The first step in my coding process was conducting open coding on all my interviews, which was done primarily via line-by-line coding. While it varied if, two, or multiple lines were used at once, these chucks of text were coded by asking myself “what is this passage about” (Charmaz, 2006).
Rather than being merely descriptive (Punch, 2013), I sought to stay close to the data and preserve
the actions and terms articulated by my informants (Gioia et al., 2013), as gerunds would provide me a better understanding of the processes at play (Charmaz, 2006). Advantages of line-by-line coding includes gaining valuable distance from preconceptions and taken for granted assumptions so that it becomes more natural to see the codes and categories emerging from the data (Charmaz, 2006). As the coding progressed, the number of repeated and unique open codes initially grew rapidly, but then slowed down as I approached the 9th interview. The progress of how many codes each interview yielded was written down in my memos and can be seen in table 4.3 below.
Table 4.3
4.3.2.1 Sampling Saturation
To determine if data saturation was reach in this study, I purposefully kept track of the number of unique codes and singleton codes2 each new interview produced. While the total number of unique codes continued to rise, the 10th interview saw a drop in the number of singleton codes (marked with bold), indicating that the data from the later informants was beginning to mirror something said by earlier informants. Having merged my codes of the 10th interview through focused coding (see forthcoming section) I conducted two additional interviews to verify to what extend saturation had been reached. Through open-coding became evident that nearly all the data from the past two interviews matched existing codes, which led to the conclusion that the desired degree of saturation had been reached.
Interview nr: Number of unique codes Number of singleton codes
2 147 105
3 213 147
4 270 122
5 287 123
6 323 129
7 352 135
8 407 163
9 447 176
10 472 163
11 485 157
12 492 150
4.3.2.2 Reflections on my open coding
My memos further showed that I felt naturally predisposed to use codes that had been used recently and that I was aware of a potential unconscious bias against codes that that hadn’t featured for a while.
Furthermore, as the software used provided an updated overview of all my codes, I was aware of the potential tendency to code for the most prolific codes more frequently. To overcome these biases, I actively attempted to code for everything that was relevant and to use new codes when in doubt, which simultaneously helped fitting the code to the data and rather than the other way around (Charmaz, 2006; Punch, 2013).
4.3.3 Focused coding
Having concluded the open coding, the second major phase was focused coding, where I identified the most significant and frequent codes that had emerged in the first phase (Charmaz, 2006). It was also in this step that the open codes which were similar got merged, which after an iterative process of comparing data with data decreased the number of codes to 125. Throughout the process, these codes were used to sift through the data once again which helped see which codes could emerge as potential 2nd order concepts and which codes were less relevant (Charmaz, 2006).
Using the Nvivo software, I started by exploring the codes with the fewest references and comparing both the code and the data it represented with other codes in my data set. Sometimes it meant merging codes with only one reference each and at other times the unique codes were merged with more prolific codes. This resulted in the merging of “having a narrow focus” into “having a myopic focus”, two similar codes that likely were coded separately to overcome the bias mentioned in the previous section. However, at other times I encountered similar codes that had important distinctions. Rather than simply merging these codes, I created a new top-level code which integrated both the codes, which allowed me to unite them without losing their distinctions. The codes
“questioning the current paradigm”, “openness to new ideas and possibilities”, “constantly trying to develop new approaches and having an innovative mentality” were united in the top-level code
“Rethinking and reflecting”.
4.3.4 Axial coding
The third major step in my grounded theory approach was trying to bring the data back again in a coherent but more abstract way than before, which builds on Corbin And Strauss (1997) idea of Axial coding. Here I looked at how different codes were interlinked, both at a more abstract level than in
the focused coding, but also in relation to the 3rd order concepts that slowly were emerging from the data (Charmaz, 2006). The distinction between axial coding and selective coding is here presented for theoretical clarity, but in practice they were hard to separate as I moved between them iteratively.
Having stayed close to the informant’s terminology in the previous steps, this phase moves towards rendering the 2nd order concepts more abstract by treating myself as a knowledge agent that seeks to answer the question “what is going on here”? (Gioia et al., 2013). A code that emerged was “Authentic commitment”, which consisted of “staying true to values”, “the importance of having top-level commitment”, “highlighting the underlying intention” and “embodiment”, which accumulated to 49 references across 10 informants. Furthermore, it was also clear that the code “authentic commitment”
frequently appeared in relation to the focused code “being purpose driven” and “the need for integrating sustainability into the organizational core”, indicating the emergence of a potential core category.
4.3.5 Selective coding
The final step in the coding process is selective coding, where possible relations between the categories are developed, leading to what can be considered core categories or 3rd order concepts (Charmaz, 2006; Gioia et al., 2013). This was done by comparing data with data, data with codes and data with emerging concepts (Gioia et al., 2013; Langley & Abdallah, 2011). These core categories were not necessarily obvious in the data (Punch, 2013), but were inferred from the data inductively, so that like any other part of GT, they earn their way into the emerging theory (Charmaz, 2006).
Besides synthesising the central points in the data and showing how they may be related (Langley &
Abdallah, 2011), the selective codes move the analytical story in a theoretical direction (Charmaz, 2006).
After analysing and comparing the 20 different 2nd order concepts that emerged in the axial coding with each other, four core categories ultimately emerged from the data. However, once I began writing the findings section it became evident that one of these categories, could successfully be integrated into the remaining three categories, a possibility that emerged by comparing the data of the different 2nd order concepts in a new way. Figure 2 shows the overall process from open coding to selective coding and how the different 2nd order concepts link to the final three core categories.
Figure 2: Data structure showing the coding process