Section 3. 1 provides a brief overview of the qualitative, interpretive approach taken in this study 2 outlines the relevance of a feminist perspective,
3.4 The Research Process
3.4.4 Data Analysis
As explained above, I see both data collection and data analysis as iterative, reflexive processes, mutually dependent, and enriching. However, alongside this implicit and on going process of analysis and refinement, the qualitative researcher must adopt a formal procedure for data analysis. My approach is based on a synthesis of procedures, outlined by Fielding (1993), Hammersley and Atkinson (1995), Dey (1993) and Musson (1994), including:
• making notes and memos immediately following interviews;
• transcribing the interviews and highlighting emerging themes and issues; • creating and assigning codes;
• "splitting and splicing" of categories; • making links in the coded material;
• examining the data in light of existing theory and constructing new theoretical frameworks for understanding.
In what follows, each of these steps will be described with reference to specific examples. It must be noted, though, that although the above list conveys a sense of linearity and clarity, the actual process of data analysis was characterised by significant ambiguity.
Notes and memos
Immediately following interviews, I recorded my initial impressions (on tape). In
addition to notes on the content of the interviews, reference was made to its context -
that is, issues concerning the respondent's work/home environment, observed
relationships with colleagues and family members, and personal factors such as the extent to which the respondents appeared to be at ease/anxious, etc. The purpose of these was to highlight issues which were thought to be significant, but which would not necessarily emerge in the interview text itself. Themes and issues noted at this early stage could then be followed-up later on in the process. Such memos contributed to a set of on-
going notes on the analysis, records which proved to be useful, both in terms of managing a potentially unwieldy process, and also in terms of the development of my understanding.
Transcribing the interviews and highlighting emerging themes and issues
Because I saw the analysis as permeating each stage of the research process, I felt that it was important to transcribe all the interviews myself. Although this was extremely laborious, it was very useful analytically, in so far as it forced me to "immerse" myself in the data. In addition to transcribing the texts, I also transcribed the memos noted above so that the interviews and my observations could be considered together. Here again, notes were made on emerging themes and significant issues which served^s the basis upon which analytical codes were constructed.
Creating and assigning codes
There is a considerable literature on the creation of analytical codes, or categories (see for example Glaser and Strauss, 1967; Miles and Huberman, 1984; Hammersley and Atkinson, 1995; Dey, 1993). Dey suggests that:
"Creating categories is both a conceptual and empirical challenge; categories must be
\grounded' conceptually and empirically... We could say that categories must have two
aspects, an internal aspect - they must be meaningful in relation to the data - and an external aspect - they must be meaningful in relation to other categories" (Dey, 1993, p. 96-97).
Consistent with Dey's approach, initial categories were constructed on the basis of the themes and issues emerging from the data thus far, together with existing theoretical perspectives. What is important, according to Dey, is the flexibility of these initial codes: "Categorisation o f the data requires a dialectic to develop between categories and ideas. Generating and developing categories is a process in which one moves backwards and forwards between the two. It is this interaction o f category and data
which is crucial to the generation o f a category set. To try to generate categories in the
absence o f both these resources would be premature " (p. 98-99).
I used NUD.IST software to facilitate the coding process and found it very useful. As suggested in the NUD.IST manual, I initially devised a skeletal list of categories, and began coding one document at a time, each time reviewing and revising my "category set" in light of the data. Here again, the notion of "progressive focusing" (Hammersley and Atkinson, 1995, p. 206) aptly describes the process. Whereas at the outset
categories were defined very loosely, as the analysis progressed, these definitions became more specific, and more exclusive. A valuable aspect of the software is its facility for recording a project's evolution. Through the use of memos, which are always easily accessed, I was able to trace the way in which a particular category, and my thinking about that category, developed.
In section 3.2 on reliability and validity, I briefly discussed the value of collaborative research . In spite of its benefits, though, I explained how in the context of PhD research it can be problematic. While my project did not involve formal collaboration in the sense described by Reason and Rowan (1981), Fisher et. al. (1986) and King (1994), in the creation and definition of categories, and the classification of data, I found informal feedback from supervisors and colleagues invaluable.
"Splitting" and "splicing" of categories
Dey explains how in the process of constructing and assigning categories, we re-focus our analysis: "this shift in focus has been described as a 'recontextualisation of the data' (Tesch, 1990), as it can now be viewed in the context of our own categories rather than in its original context" (Dey, p. 129). Thus it is the categories themselves that become central to our thinking.
Dey uses the terms "splitting" to describe the process by which we sub-divide the data assigned to a particular category in order to further refine and differentiate that category. "Splicing", in contrast, refers to the joining and "interweaving" of categories so as to deepen our understanding. Thus, "we split categories in a search for greater resolution
and detail and splice them in a search for greater integration and scope" (p. 139).
However, whereas Dey describes the processes of splitting and splicing as following the
creation and assignation of categories, in my experience these processes happened simultaneously. For example, I initially created a category which I called "pre-transition job", referring, as the name suggests, to the organisations in which respondents worked
prior to the move to self-employment. However, after working through several transcripts it became clear that this category was far too broad - in order to be
analytically useful it needed to be split and refined. I therefore sub-divided the category and re-assigned its data at that point, and with each subsequent transcript further
examined the category to make sure it still "worked". At the same time, there were times when I realised that the categories I had created were too narrow, resulting in an overly fragmented analysis. In such cases categories were spliced in order to achieve a more holistic picture.
As the "category set" became established, I was able to define each category with greater accuracy and precision. Having completed two-thirds of the transcripts my analytical framework was basically set, with only minor adjustments in the last eight documents. Making links in the coded material
While categorisation enables us to closely examine the data and to explore similarities and differences within it, as suggested above it can result in fragmentation and a lack of integration - it can serve to shatter the "big picture". In the discussion of creating categories above, it was noted that not only must categories make sense in terms of the data, but they must also make sense in terms of each other. Thus, underpinning this
approach is a view of these categories as relational. Dey insists that although the
process of categorisation leads us deeper and deeper into each category, it is at the same time crucial that the researcher does not lose sight of these relationships, of how the categories interact. Thus he argues that "we need to link data as well as categorise it" (p. 152).
In seeking to establish such links, the "career” and "relational" models, described in section 3.4.3 as illuminating both diachronic and synchronic perspectives, are particularly
useful. It would be possible to focus the analysis on the discrete categories themselves; however, this would result in a failure to appreciate the ways in which these categories interact. Whereas such a narrow focus would have provided a limited, and very static picture of the data, these diachronic and synchronic perspectives allow for more complex, more dynamic analyses. For example, seeking to understand women's
decisions to leave their organisations could simply involve a detailed examination of the data coded in the "pre-transition job" category. However, such an analysis would have been partial and insufficient. A more adequate analysis would examine the links between that category and those focusing on relationships, both at home and at work, as well as the data on women's goals and aspirations. What is essential, then, is not only an examination of the most minute details of the data, but also of the relationships
embedded within these data. ^
Using discourse as an analytical tool
This study, like much qualitative research, is based on talk. Of primary interest to researchers working qualitatively, therefore, is to find ways of understanding this talk which is our data. In addition to the processes of coding, splitting and splicing, outlined above, the concept of discourse can be used to highlight patterns of thought and
understanding which are embedded within and articulated through respondents’ accounts. The definition of discourse which is being used in this study was detailed in section 2.1.4. Central to that definition is the incorporation, within discourse, of both structural and agentic dimensions of experience. Indeed, it has been suggested that it is though discourse that this dynamic relationship is played out (Bahktin, 1981; Fairclough,
1992). Also fundamental to that definition is the role that discourse plays in the construction of understanding.
The process of identifying discourses is a subject of considerable academic debate (Fairclough, 1992; Casey, 1993; Cohen and Musson, 1996;), typically involving an examination of texts on both macro and micro levels. As an analytical tool, discourses can be used in a number of ways (Cohen and Musson, 1996), depending on the focus of the particular study. One could track the way in which a single discourse is constructed within a particular account, or by diverse people and groups. For example, in her study
of women teachers Casey (1993) found that permeating the stories of a group of Catholic respondents was a religious discourse, articulated in a variety of (sometimes ambiguous) ways. Casey felt that it was this discourse which gave these stories their underlying logic, and which provided her, as a researcher, with a conceptual tool for understanding.
Conversely, the dynamic relationship between discourses can also be explored. For
example, in her study of how general medical practitioners interpreted and acted upon reforms imposed upon them by the NHS, Musson (1994) identified two central
discourses: a business discourse and a clinical discourse. A consideration of the ways in which these colliding/overlapping/competing discourses were constructed and articulated by the doctors in her study provided important insights into their perceptipns of their
working contexts, and of themselves as GP’s. It is in this relational sense that the
concept of discourse as an “analytical tool” is being used in this study. Thus, this process of identifying and the examining the relationship between key discourses acts as the disclosing table which reveals how the women in the study come to construct their personal and professional realities, and thus make sense of their move from employment to self-employment.
Examining the data in light of existing theory and constructing new theoretical frameworks for understanding.
The researcher's understanding of the complexity of the categories themselves, as well as their relationships with other categories, develops as the analytical process continues. However, Hammersley and Atkinson (1995) argue that such development is rarely characterised as pure induction: theoretical ideas, as well as common-sense
understandings also contribute to the analysis. As regards the former, they suggest that: "Where a category forms part o f a typology or model developed by others, however loosely constructed, relations with other categories may be implied that can be tentatively explored in the data" (p. 212).
Existing typologies can thus be "tested" in relation to the data at hand. Hammersley and Atkinson suggest, however, that most models are not sufficiently robust to be applied in this way. In their view, "the process of testing requires considerable further development of the theory or explanation" (p. 214). This has certainly been so in the case of this study. For example, section 2.2.4 discusses the potential usefulness of a typology used to explain the move to self-employment, put forward by Stanworth and Stanworth (1995). In section 6.2, which focuses on respondents' decisions to become self- employed, this typology is "tested" in relation to the data generated in this study (by "tested" I mean that it was applied to the accounts of each of the twenty-four
respondents). As reported in detail in that section, where it appeared to "fit" for a small number of women, problems emerged when applied to the majority. Questions were then asked about what those few had in common, which differentiated them from most of the other respondents. Thus a dialogue ensued between the typology and the data, and between the accounts of various groups of respondents themselves, eventually resulting in a much clearer understanding of those aspects of the typology which were useful and should be retained, and those which needed further refinement and revision.
Lofland (1970) maintains that in the development of typologies, and new theoretical frameworks for understanding, much ethnographic work suffers from "analytic
interruptus"! That is, he suggests that analysts "'fail to follow through to implied logical conclusion... to reach [the] initially implied climax'" (p. 42)! Lofland's approach is systematic and painstaking, and involves:
• a thorough examination of all the data generated on a particular issue; • the "teasing out" of variations, discontinuing and exceptional cases; • their classification into an "articulate set of types"; and
• an orderly presentation of the resulting typology or model.
The approach to analysis and theory-development undertaken in this study broadly follows these recommendations. This constant interplay between the data and the emerging analysis is rigorous and ensures a high level of attention to detail, elucidating both the similarities and differences between different respondents' accounts, and within individual stories.