The assignment of short interpretive labels to segments of qualitative data — interview transcripts, field notes, documents, images — to support pattern detection, retrieval, and conceptual development. Coding is the linkage between raw data and analytic claims; the choice of coding method follows from the methodology rather than the other way round.
Saldaña (2021) organises coding methods into two cycles. First-cycle coding operates on raw data and produces an initial pool of codes. Methods at this cycle include descriptive coding (topic labels), in vivo coding (using participants' own words), process coding (gerund-form actions), values coding, emotion coding, versus coding, evaluation coding, and structural coding among many others. Second-cycle coding works on the first-cycle output to consolidate, restructure, and theorise. Methods include pattern coding, focused coding, axial coding, theoretical coding, elaborative coding, and longitudinal coding. Saldaña's manual catalogues thirty-five coding methods across the two cycles.
Grounded Theory in the Strauss-Corbin tradition uses open, axial, and selective coding mapped to a coding paradigm of conditions, actions/interactions, and consequences. Glaserian grounded theory begins with substantive coding (open, then selective) and moves to theoretical coding using "coding families." Thematic Analysis in Braun and Clarke's reflexive form treats codes as analytic units developed across a recursive engagement with the dataset, with themes constructed at a later stage rather than identified at the coding stage. Framework analysis, template analysis, and content analysis use a structured codebook applied — and refined — across the corpus, often by multiple coders.
Coding decisions require an audit trail: definitions, examples, and decisions about boundary cases recorded in a codebook or in memos. Software (NVivo, MAXQDA, Atlas.ti, Dedoose) supports retrieval and matrix queries but does not perform the analytic work. Inter-rater reliability is appropriate to coding-reliability and codebook approaches; reflexive approaches reject it as a category error.