Open Coding — Initial Qualitative Coding

Open coding and thematic analysis both begin with an inductive read of qualitative data, but their conceptual logic differs. Open coding, rooted in grounded theory, aims to label every discrete incident or idea in the data and then compare labels to build categories that will feed a theory. Thematic analysis looks for patterns of meaning (themes) across the dataset as a whole and does not require the granular incident-by-incident pass that open coding entails. In practice, many thematic analysis studies use open-coding-style initial reading as their first step, but the two methods differ in their ultimate goals and in how categories are developed.

No. Open coding is a technique that originated in grounded theory but is now applied in many qualitative traditions as an inductive first-pass labelling procedure. You can use open coding as the initial phase of a thematic analysis, a content analysis, or a narrative inquiry without committing to the full grounded theory methodology. However, if you do use grounded theory, open coding is not optional — it is the obligatory first stage of the three-part coding sequence.

In-vivo codes use the participant's own words as the code label — for example, labelling a segment 'flying under the radar' because that is exactly how the participant described their strategy. They are valuable when the participant's language is analytically significant, captures a concept the analyst's vocabulary might flatten, or when preserving emic (insider) meaning is a priority. In-vivo codes are especially emphasised in Charmaz's constructivist grounded theory. Use them selectively alongside analyst-constructed codes rather than exclusively.

There is no fixed ceiling, but if you are generating a unique code for almost every sentence, constant comparison is not doing its work. A productive open-coding pass on a set of 20 interviews typically yields between 80 and 300 initial codes before consolidation into categories. If you have several hundred codes and cannot see clusters forming, pause and run a deliberate comparison session — group similar codes, rename categories, and collapse codes that refer to the same phenomenon.

Yes, and doing so strengthens credibility. After independent coding, the analysts meet to compare their codes segment by segment, discuss disagreements, and develop a shared codebook through consensus. This process is sometimes quantified with a percentage agreement or Cohen's kappa statistic, though many qualitative methodologists prefer to treat disagreement as analytically productive — a sign of complexity in the data — rather than as error to be eliminated.