Data management and analysis

The recorded interviews were transcribed verbatim by experienced transcribers and were checked for accuracy by four research team members (SM, DC, ED and CM). The interview transcripts were then translated from Kiswahili into English by a professional translator and the translations checked for accuracy by the Principal Investigator (SM). The first four authors (SM, DC, ED and CM) each read between five and 10 transcripts to familiarize themselves with the data. Two members of the research team (SM and DC) developed a code manual based on the objectives of the study and conceptual framework. The codebook was shared for review with senior researchers (ZS and KR). Using NVivo10 qualitative data analysis software [18], three members of the research team (SM, DC and CM) independently coded the first five interviews to develop consistency. Thereafter, SM and DC coded the remaining transcripts. New codes that emerged during the coding process were added with consensus from all research team members. Saturation was achieved when no more codes emerged from the data. Key themes were then independently identified by each of the coders and organized by level of respondent to facilitate comparisons. The themes were discussed by the researchers. Finally, two researchers (SM and DC) identified representative quotations for each key theme, and obtained consensus from all team members. The study thus used deductive and inductive methods to generate the themes [19]. The key dimensions from Walt & Gilson’s policy analysis framework—actors, process, context and content—informed the deductive approach to analysis and the format for reporting our findings in the following section.