Qualitative Data

Qualitative data is information that describes qualities, categories, or meanings rather than amounts — interview transcripts, open-ended survey answers, field notes, photographs, product reviews, the color of a flower, the brand of a phone. Its defining feature is that it records what kind or how something is, so the natural operations are grouping, comparing, and interpreting rather than adding and averaging, and the first question asked of it is "what themes or types are present?" rather than "how large is the value?"

What separates qualitative data from its quantitative counterpart is the form of the raw record, not the topic: a qualitative record is expressed in words, images, or labels, and any numbers must be created afterward by a deliberate act of coding. A free-text customer complaint is qualitative; counting how many complaints mention "battery" is a quantification of it. This is why the term describes a stage of the data rather than a fixed property of the subject: the same hospital stay yields qualitative data when a patient narrates it and quantitative data when the length of stay is logged in days. The tidy label-only subset of qualitative data — responses that already fall into a fixed menu such as blood type or country — is called categorical data, and is the part that standard statistical tables and tests can handle directly, while rich unstructured material needs interpretive methods first.

Rigor in qualitative work comes from systematic procedure, not from sample size. Methods such as thematic analysis, content analysis, and grounded theory (introduced by Barney Glaser and Anselm Strauss in 1967) tag passages with codes, merge codes into themes, and test those themes against new material until little new emerges, a stopping point called saturation. Reliability is checked by having two coders label the same material and measuring agreement, commonly with Cohen's kappa (proposed by Jacob Cohen in 1960), which discounts the agreement expected by chance. Qualitative studies typically use purposive samples of tens of participants because the goal is depth and explanation of mechanisms, and generalizing a percentage to a population is not the aim.

Qualitative data is the foundation of exploratory and explanatory work: it generates hypotheses that numbers can later test and explains why a pattern in the numbers exists. Product teams read support tickets and run user interviews to learn why users churn; epidemiologists use focus groups to understand why a health campaign failed in one community; historians and anthropologists work almost entirely with such evidence. In data science, the explosion of text, image, and audio data has made qualitative material machine-processable: sentiment classification, topic modeling, and embedding models convert words into vectors, turning qualitative records into quantitative features at scale. That conversion is useful but lossy, and the labels a model emits inherit whatever the coding scheme and training data chose to emphasize.

Several misconceptions recur: that qualitative means unscientific, that it means subjective opinion, and that it is the same thing as categorical data. Disciplined qualitative research is replicable in its procedures and transparent about its coding rules, so it is not unscientific merely because it avoids p-values. It is also not limited to opinions, since observed behavior recorded in notes is equally qualitative. And while all categorical data is qualitative, the reverse fails: a paragraph of free text is qualitative without being a category at all. A further pitfall is mistaking numbers for quantities: a jersey number or a postal code is a numeral attached to a label, and its qualitative nature is revealed by the fact that averaging it is meaningless.

Read qualitative data as evidence about what and why — the nature, meaning, and context of what happened — whose power lies in careful interpretation and coding, and which turns into countable quantity only at the moment a researcher decides how to categorize it.