6. Methodology 145
6.2. Expert interviews 151
6.2.3. Data processing and analysis 157
6.2.3.3. Approach for the analysis and evaluation of expert interviews conducted
On the basis of Mayring´s qualitative content analysis and Boeije´s approach to constant comparative analysis described above, as well as taking into account work by Meuser/Nagel and Schmidt, an evaluation/analysis strategy was developed that combines the evaluation/analysis procedures of the scholars mentioned (see Meuser/Nagel 2005, Boeije 2002 and Schmidt 2004). Not a detailed single case study analysis is the aim, but the possibility of interpreting and generalising commonly shared views, interpretations, areas of knowledge and constructions of all experts (see Meuser/Nagel 2005: 80). Also, the aim is to work out the differences among the individual perceptions of the interviewees in order to identify distinct patterns and concepts.
217 Boeije 2002, p. 406.
161 In the following, four steps or phases (transcription; paraphrase and unitisation; coding and categorisation; conceptualisation and answering of research questions) for the analysis of the
expert interviews conducted in Uganda are presented in more details:219
(1) Transcription
The basis for the transcription is a tape-recorded interview. In the case of expert or key informant interviews, no complete analysis of the material is done, since pauses/intervals, pitch of the voice and other non-verbal or para-linguistic elements are not part of the analysis. The degree of transcription depends on the quality of the discourse.
In this study, a total of 20 interviews were fully transcribed, totalling 193 pages for analysis. Of these, 18 interviews were conducted and transcribed in English, two in German.
(2) Paraphrase and unitisation
The paraphrase is the first step towards the condensation of the material: the statements of the experts are reproduced accurately and in chronological sequence according to the original text. A good paraphrase is characterised by a non-selective relationship to the discussed issues or topics; it is entirely protocolary and directed to the content. A paraphrase is necessary if the transcribed material is very extensive and needs to be reduced to a manageable size and does not have any major implications if the quality of the text is not part of the analysis. Paraphrasing is, therefore, not a necessary step for the unitisation of the material. The research question determines whether words, sentences or text chunks are chosen as units on which coding and analysis are based.
In this study, so-called “thought units”, “sense units” or “units of meaning”, which comprise one idea being communicated, are used as a basis for coding and analysis.
(3) Coding and categorisation
Categorisation is a process that involves structuring and condensing data by grouping the qualitative material in theoretical insightful ways. For this step, the two methods of qualitative content analysis and constant comparative analysis are relevant.
In this study, the following procedure was used:
a) the coding process was started by reading through each of the (paraphrased) transcripts and attributing a code to “units of meaning” or “thought units” (sentences, paragraphs or sections). These codes can be seen as headlines that represent a theme, topic or idea with which each part of the data is associated. The terminology of the respondents should be taken up, while at the same time the headings should give an overview on the main statements of the key informants. Passages of transcripts were given no code, one code or several codes. The codes were written on the hard copy of each transcript next to the related section. The codes were also separately written down on another paper, generating an unordered list of codes. Having coded the first transcript, all other transcripts were coded with the first one in mind, adding new codes wherever necessary. The researcher should not simply use the questions asked as a basis for formulating the headings, but should pay attention to use the
162 phrases or wordings that the experts themselves used throughout the interview. The accurate and repeated reading of the texts is very time-consuming, but it is a necessary step in order to ensure that passages are not rashly associated with certain topics or headings and that passages whose relation to the question is not clear at first sight are overlooked. The repeated reading of the single transcripts also helps to check the interviews for consistency by comparing statements that were made at one point of the interview to other remarks made earlier or further on. The researcher has to take care not to fit the material into his/her previously developed theoretical assumptions: there is a tendency to see some fitting passages at first sight and overlook those that do not fit the personal expectation, a problem that can be controlled by going through the material over and over again. The application of the open coding procedure merely helps to develop codes inductively out of the transcripts.
b) After coding each transcript, the documents were divided into broad/rough sections or chapters. Each section from each interview was then cut and pasted and put into a new document, a computer file with the provisional title given to this section, including the number of the interviewee and the line numbers from the original transcript. In this way, the extracted sections could always be traced back to the original transcript in case there was a need to look into the contextual details. This was done to facilitate the data analysis and
manage complexity, since it was done manually220. The sections drawn were the following: (1)
description of the organisation and its main activities; (2) PSD policy of the Government; (3) issues related to MSEs (definition, challenges and strengths, associations, etc.); (4) business linkages and the value chains approach in general (the usefulness and potential, examples, etc.); (5) issues related to the development or promotion of business linkages and value chains (issues related to the operational or implementation level).
c) Each section or chapter was now read subsequently, first one by one per interviewee. In some cases, the coding was further detailed. In a second step, interviews were compared pair wise, i.e. the first with the second interview, the second with the third and so on. Coded sections were compared with other segments to ensure consistency of application of the codes, but mainly to identify differences and similarities among the interviews, starting to work out patterns.
d) Once coding was completed, the codes that had common elements were merged together to form preliminary categories. Files were created for each category, containing copies of the codes that had been merged to form the category. Some codes were placed in more than one category. Further comparative steps were undertaken, now within each category, namely comparison of short series of interviews. First, comparison within each series of interviews was undertaken, then among the different series of interviews. In this study, interview series were once defined according to the numbering of the interviews (4 in total, i.e. interviews 1 to 5, interviews 6 to 10, etc.) as well as according to the represented type of organisation (3 in total, i.e. public institutions, donor agencies, private sector and NGOs). The aim was to find
220 The term “manually” refers to the fact that no computer programme was applied in order to facilitate content
163 out whether the patterns emerging from one series differed from the patterns emerging from another series and if so, what conclusions could be drawn out of this fact. All these comparative steps helped detail the preliminary categories more and more. Part of this was a simple quantitative analysis: the number of mentions in relation to each code was counted (and
specified for the three groups of stakeholders – government, donors and private sector).221
(4) Conceptualisation and answering of research question(s)
Once all comparative steps are undertaken and the categories defined, the conceptualisation stage starts. A separation from the respondents’ terminology takes place: findings (similarities and differences, but also relations among categories) are described in a scientific language comparing them to relevant theories and other empirical studies. This can also imply a quantitative compilation of the results of the coding process: frequencies and distributions in the material can be described and categories can be put in relation to each other using cross-tables. Quantitative compilations of the material mainly have the function of preparing the ground for further quantitative analysis: they may point to interrelations that can be analysed in more detail through quantitative (correlation, factor analysis) or qualitative techniques (for example detailed case analysis). Each research question can be allocated one or more categories that contribute to answering it; each category can contribute to answering one or more research questions.
A number of experts provided copies of relevant reports, policy documents, research studies and evaluations – the key findings from these were considered as background for the analysis of the transcribed interview material, along with other relevant literature gathered.