A topic in the Open Knowledge Graph — a free, open map of 15,290 topics and the order to learn them in.
Qualitative data analysis involves reading, coding, and interpreting text. Codes are labels capturing meaning; themes aggregate codes into patterns. Inductive coding emerges from data; deductive coding applies predetermined categories. Inter-coder reliability and audit trails enhance credibility. Analysis continues until saturation (no new themes emerge).
Code an interview transcript using inductive and deductive approaches. Compare your codes with a colleague and discuss discrepancies. Extract and report themes with illustrative quotes from raw data.
Your prerequisite study of qualitative interview methods showed you how to collect rich textual data — interview transcripts, observation notes, documents. Now the question is: what do you do with it? Qualitative data analysis is the systematic process of transforming raw text into interpreted meaning. Unlike quantitative analysis, which applies predetermined statistical operations, qualitative analysis involves iterative reading, labeling, comparing, and abstracting. The result is not a p-value but a set of themes or categories that capture the structure of meaning in the data.
The first step is coding — attaching labels to segments of text that capture what is happening in that segment. Codes can be applied at different levels of abstraction. A descriptive code labels what a segment is about ("participant describes fear of failure"); an interpretive code offers a higher-level meaning ("perfectionism as coping mechanism"). In inductive (bottom-up) coding, you derive codes directly from the data without pre-existing categories, reading and re-reading until patterns emerge. In deductive (top-down) coding, you begin with a theoretical framework or predetermined coding scheme and apply those categories to the data. Most analyses blend both: beginning inductively to capture what is genuinely new, then applying theoretical concepts as organizing frameworks. The choice reflects your epistemological stance — grounded theory researchers resist premature theory-imposition; framework analysis researchers use theory as scaffolding from the start.
Thematic analysis is the most widely used method for moving from codes to findings. After coding, you group related codes into higher-order themes — clusters that capture a coherent aspect of participants' experience or meaning-making. A theme is not simply a topic that appeared frequently; it is a pattern of meaning that illuminates something about the research question. This is why the misconception that "frequency equals importance" is problematic: a theme can be analytically central even if only a few participants voiced it, if it represents a structurally significant aspect of the phenomenon. Conversely, a code applied 300 times may be background noise rather than a meaningful finding. Interpretation, not counting, drives thematic analysis.
Inter-coder reliability is the mechanism that addresses the legitimate concern about subjectivity. Two or more coders independently code a subset of the data using the same codebook, then compare their assignments. Agreement can be quantified using percentage agreement or Cohen's kappa (which corrects for chance agreement). High agreement (kappa ≥ 0.7 is a common benchmark) demonstrates that the codes have been defined clearly enough that different researchers apply them consistently — not that subjective judgment has been eliminated, but that it has been rendered transparent and reproducible. When coders disagree, the discussion of disagreements often improves the conceptual clarity of the codes themselves, making the process generative rather than merely corrective.
Saturation is the criterion for knowing when to stop collecting data: the point at which new interviews or observations yield no new codes or themes. This is a functional rather than mathematical criterion — it has nothing to do with sample size per se, which is why the misconception about minimum sample sizes is misleading. Saturation depends on the heterogeneity of the phenomenon and the specificity of the research question. A homogeneous population (experienced teachers in one school) may saturate at 10 interviews; a heterogeneous phenomenon (experiences of grief across cultures and relationships) may require many more. The rigorous approach is to document your saturation judgment — to explicitly note when the last three or four interviews added no new material — and to report this as part of methodological transparency, not as a magic number.