2.4 Data analysis

We used an inductive, thematic approach to analyzing the qualitative data. Our iterative analytic strategy drew on contemporary grounded theory methods, described by Charmaz as “systematic, yet flexible guidelines” for qualitative analysis and theory-building [32]. The co-PIs agreed on a set of initial codes designed to reflect a priori domains of interest from the interview guides and exposure to the raw transcripts and field notes. The lead author (EH), who has over 10 years of experience working in Kenya, and a Kenya-based expert qualitative data analyst (EC) coded 5 IDIs in parallel, after which they compared code application and meaning, adding new codes as needed to identify emerging themes. After a second round of parallel coding, investigators constructed a final codebook for both IDIs and FGDs, and re-coded or coded all remaining transcripts. Two-thirds of transcripts were double-coded, and investigators met regularly to discuss coding discrepancies and emerging themes. The analytic team identified the most significant codes, grouped similar and contrasting excerpts within and between codes, and wrote analytic summaries for each major theme. Investigators conducted all coding in Dedoose (Version 8.0.35, SocioCultural Research Consultants, LLC, www.dedoose.com), an online qualitative analysis program facilitating data management and collaborative coding. The conceptual model (Fig 2) was developed during the analytic process in order to organize and interpret themes emerging from the data.