Coding and analysis

100 Madden (2010:141), coding is a dual exercise of both ethnographic facts and the ethnographer's choice.

Coding qualitative material is thus a matter of indexing concrete events in the field as well as organizing the interpretive aspects of the material (see also Kouritzin 2002). The code book should not be

considered a proof or a source of any truth telling, and the coding process should not be viewed as a test that verifies any reliability of the analyses.

What I will do in the following is to account for how the qualitative material for this project was coded and interpreted in order to make my choices and understandings transparent for the readers. Readers can then make their own validity judgments as to whether they find my analyses convincing.

101 3.9.1. First analytical step: transcribing

Although transcribing is a lengthy engagement, I find that much can be gained from doing our own transcriptions, as it offers an opportunity to relive and internalize the interviews by writing them out.

Therefore, I conducted my own transcriptions of all recorded interviews.

Although I transcribed a few interviews during the fieldwork, the largest bulk of the interviews were transcribed after completing the fieldwork. In hindsight, I might have benefitted from transcribing more interviews on a continuous basis, as the transcription allowed me to assume an observational role as a third party in the interview situation where I could pay attention to what was being said without having to consider the next question; how to make the interviewee elaborate, or how to explore their

perceptions even further. It also offered important insights into my own interviewing techniques that could have been refined and improved along the way.

On the other hand, doing the transcriptions in bulk allowed me to re-live the interviews in the context of one another and to identify similarities and differences that I had not noted while conducting the interviews, some of which were many months apart. This is precisely why I view the transcription process as the first analytical step and – ultimately – the first step in the coding phase, as this exercise greatly informed my coding process.

3.9.2. Second analytical step: familiarization and first order coding

I have used the qualitative data processing software NVivo to code my ethnographic material (pictures, field notes, documents, observations, intranet stories, emails, slides and interview transcripts). As I undertook the coding process abductively (cf. Mantere and Ketokivi 2013; Reichertz 2014), I did not create the codebook beforehand. Rather, the codes were formulated as I reviewed the material. The purpose of this approach was to allow the ethnographic to define the topics to be coded rather than forcing the empirical material into a pre-defined codebook. Of course, I am not suggesting that this is a more objective approach, since it is I as a subjective actor who has generated the empirical material.

Nevertheless, the codebook is an attempt to approach the various sources on their own terms rather than immediately assigning theoretical concepts to the material (although, I am aware that I may have done so already by the way I have performed the fieldwork, selected events to study and chosen questions to ask). The downside to this more inductive coding approach is that the entire material will not be coded according to the same codebook. In order to mitigate this disadvantage, however, before starting to use NVivo, I read through the entire data material and noted down initial observations and ideas, resembling the process that Clarke et al. call ‘familiarization’ (2015:230–32). Through

102 familiarization, the researcher engages with the dataset in a thorough way before commencing the coding phase, and it involves listening to all recordings and reading all notes and material.

After having coded half the material in NVivo, I reached a stage where I did not add new codes but could reuse the existing ones. In order to ensure that the entire dataset had been coded using the full codebook, I recoded the first half of the material again upon completion of the first coding process. I then reviewed all coded material for each code so as to ensure that the content did indeed reflect the code I had assigned to it and my final understanding of it.

With regards to the type of coding employed, as described by Clarke et al. (2015), codes can often be divided into semantic codes and latent codes. Whereas semantic codes remain at the descriptive level and simply refer to what is being said or done, latent codes are more analytical. They move beyond what is stated by the informant (Clarke et al. 2015:235). Clarke et al. recommend to carry out the latent and semantic coding simultaneously, but I have strived to follow the recommendations of Gioia et al.

(2012), (first mentioned in Gioia and Chittipeddi 1991), to divide the analytical phase into first order and second order coding. First order coding is, similar to semantic codes, a coding process with informant-centric codes, whereas second order coding is characterized by more analytical coding, similar to the latent codes mentioned by Clarke et al. (2015). The approach by Gioia et al. (2012) has a temporal separation of the two coding styles that I found helpful in my study in order not to introduce analytical concepts and interpretations too soon in the coding process.

Thus, I coded the entire dataset with semantic first order codes that were either descriptive or what Saldaña terms ‘in vivo codes’, i.e., codes that refer to statements made by informants (Saldaña 2016:4–

5). In practice, however, in order to limit the amount of codes to a somewhat manageable number, some codes started as in vivo codes but were later used as descriptive codes in those instances where informants put similar statements in different wording. For example, the code ‘Ferring philosophy is common sense’ was created as an in vivo code where the informant (and several informants thereafter) literally stated that ‘…for me, the Ferring Philosophy is common sense’. However, I broadened the scope of the code and also used it as a descriptive code for statements such as the following: ‘A lot of it is also relatively straight forward in a way. If you are a decent person, with some values and things, then there is not too much that is so surprising [about it]. (…) It’s a bit... obvious.’⁵⁰ I found that statements such as this represented a similar meaning: that the content of the philosophy is ‘common sense’. In

50 Please see Appendix 1-5 for more examples of how the empirical material has been coded.

103 order to address that some of the first order codes were connected yet separate themes, some of the first order codes have been assigned a number of sub-codes.⁵¹ Although this adds another level of complexity to the coding structure, it is also a tool for me to ensure that important nuances of the codes were retained and given individual attention. Thus, the methodology applied in this project varies from the approach presented by Gioia et al. (2012). However, as they also emphasize, their coding method should not be viewed as a template but as flexible enough to accommodate the needs of each individual study (Gioia et al. 2012:26).

Moreover, all interviews were coded with attribute codes, which essentially is to attach demographic categories to each interviewee (vocational backgrounds, country of residence, managerial levels, etc.) (cf. Saldaña 2016:82). Following the advice of Alvesson and Kärreman (2011), and in an attempt to instil some disorder in the coding process to allow for surprising insights, the material was coded randomly, not in chronological order, and I shifted between interviews, documents and field notes.

3.9.3. Third analytical step: second order coding and pattern recognition

As mentioned earlier, whereas first order coding is an attempt to structure empirical material very close to actual events and statements, second order coding introduces more analytical concepts and

categories with higher levels of abstraction. My second order coding was an iterative process. I read through the material coded under each code, attempted to cluster codes under common themes, then read through excerpts again and regrouped them until I had a structure that seemed coherent both from the perspective of each thematic cluster and with the entire data material in mind. Although coding involves some form of reduction and detachment of excerpts from their context as they are ordered into the more manageable coding units, it is of outmost importance that the researcher

maintain an overview of the material as a whole and insists on viewing each coded cluster of excerpts in relation to their context. I have strived to do so by employing this coding strategy.

Furthermore, as noted by Saldaña (2016:7), recognizing patterns is not only about recognizing similarities and commonalities but also about recognizing the patterns that lie within differences and idiosyncrasies. As Alvesson and Kärreman (2011) advise, we must not let pattern recognition simplify and restrain our analysis, as the irregularities and deviations are key aspects of the social realities that

51 The analyses found in this dissertation stem from the codes and excerpts illustrated in Appendix 1-5, and the reader will thus recognize the themes and concepts from the analyses in these codes and excerpts. However, as noted by Gioia et al. (2012), the static image of the data structure itself is not sufficient for, as the analysis demands a close familiarity with the data and an overview of it in its entirety.

104 we strive to comprehend. These anomalies or outliers must therefore be explored in relation to – and in conversations with – the patterns we recognize (Alvesson and Kärreman 2011:41–43).

In the empirical chapters that follow in this dissertation, I have strived to explore both the patterns and the outliers from these patters.

3.9.4. Fourth analytical step: Writing up

The last formal analytical step that deserves mentioning here is the leap from individual interviews and field notes to analysis that happens in the writing phase. The writing process is as much an analytical realization process as the previous steps. During write-up, insights gained from the transcription and coding phases materialize into theoretical insights and an ethnographic narrative. Thus, the writing phase consists of much more than simply ‘writing up’ the ‘data’. The reader will be able to evaluate the result of this endeavour throughout the dissertation.