79 From Critic to Inspirer 3 from 30 hours to 12-22 hours per week.46 At the time of the interviews, the medical curriculum of the Leiden University Medical Centre (LUMC) had been reformed in the decade before from a traditional teacher-centred curriculum towards a student-centred curriculum. Central educational characteristics of the curriculum were: a student-centred approach, active learning and emphasis on learning, thinking, and performing strategies rather than on only obtaining factual knowledge. The numbers of large group lectures were reduced, small group teaching sessions were introduced, and part of the study time was designated for self-study.47 Analysis The analysis of the data followed the steps of the study design. The software programme Atlas-ti was used as the main supportive tool. After immersion in the data, the first step was largely deductive: within the first twelve transcripts three team members (MO, LvS, RvdR) selected those text fragments that related to our first research question using the six levels of the onion model (Figure 3.1) as initial themes, even though we kept the possibility for new themes open. These twelve transcripts were analysed independently to ensure data dependability. For each of the six themes the descriptions as formulated by Korthagen were the starting point; demarcation rules for each theme were fine-tuned during the analysis process to enable a consistent coding. We chose to code a text fragment within the convictions theme only if it could not be placed in one of the other five themes. As a second step, since we uncovered a broad diversity of content within each of the themes, we developed, revised and refined subthemes inductively within each theme, using the procedure of constant comparison.48 Descriptions and demarcation rules with respect to the subthemes were discussed within the research team, together with example text fragments. Within each theme, we reached ‘code saturation,’49 meaning that no new subthemes emerged after we were about halfway through the analysis of all text fragments. To calculate inter-rater agreement of the coding of themes and subthemes, the first author selected the text fragments of four not yet coded interviews –a total of 55 text fragments– as meaning units, which were coded independently by two team members (MO, LvS). This resulted in an inter-rater reliability of 0.87 (Cohen’s Kappa). Consensus was reached on the text fragments that were coded differently. The remaining transcripts were analysed by the first author,
RkJQdWJsaXNoZXIy MTk4NDMw