Can qualitative data contain numbers?

Yes. Numbers may serve as category codes rather than amounts.

Can qualitative observations be counted?

Yes. Their counts and proportions are quantitative summaries of categories.

Are low, medium, and high quantitative measurements?

They are ordinarily ordinal categories. Equal distances between the levels should not be assumed without justification.

AI Glossary / Statistics and probability / Beginner

Qualitative data

Definition

Qualitative data describe types, qualities, or categories rather than measured amounts or counts. Examples include a support request's topic, a product's color, and a customer's stated reason for cancellation. Such data may be stored as words, symbols, or numerical codes. A number used only as a category label does not become a quantitative measurement simply because it looks numeric. This entry follows the statistical data-type meaning used in the supplied glossary.

What it means

Qualitative data answer questions such as what kind, which category, or what characteristic. A complaint labeled billing describes its type; the number of billing complaints is a quantitative summary calculated from those labels.

Categories can be nominal, with no inherent order, or ordinal, with a meaningful order but not necessarily equal distances between levels. Low, medium, and high severity are ordered categories, but the gap from low to medium need not equal the gap from medium to high. In qualitative research, the term also includes richer material such as interview accounts and observations, which may require interpretation beyond a predefined classification table.

How it works

Begin by defining the question and the unit being described. Decide whether a response belongs to one category, several categories, or an open-text field. Provide a codebook explaining each label, its intended meaning, and how ambiguous cases should be handled.

Collect or annotate the data consistently. Distinguish an explicit category such as no issue from missing information or an uncertain classification. If categories are assigned by people or AI, assess the labeling process and examine disagreements. An automated label is a prediction or annotation, not automatically the underlying truth.

Choose summaries that respect the scale. Frequencies, proportions, cross-tabulations, and bar charts often suit categorical data. Ordered categories can retain their ranking, but numerical averages of arbitrary codes may have no useful interpretation. For machine learning, encoding choices should preserve the intended meaning and be evaluated within the model pipeline. Maintain the original labels and mapping so that outputs remain understandable.

Three practical examples

A customer-service team records whether each request concerns billing, delivery, or account access, then compares category frequencies.
An AI evaluation assigns responses to correct, partially correct, and incorrect categories under a documented rubric, preserving their order without assuming equal spacing.
A product researcher reviews open-ended comments to identify themes and records the evidence behind each interpretation rather than reducing every comment immediately to a score.

Think of it this way

Think of labels on folders. The labels help organize what is inside, but replacing the labels with folder numbers does not make folder 30 three times as important as folder 10. The numbers may be useful addresses. They become quantities only if their meaning supports quantitative operations. Qualitative information can be organized precisely even when subtraction or averaging is inappropriate.

Worked example

In an original example, 40 support tickets receive exactly one topic label each: 18 billing, 14 delivery, and 8 account access. Their proportions are 18/40 = 45%, 14/40 = 35%, and 8/40 = 20%. These proportions summarize categorical observations; the topic itself remains qualitative.

Suppose the database codes billing as 1, delivery as 2, and account access as 3. The numerical average is (18 × 1 + 14 × 2 + 8 × 3)/40 = 70/40 = 1.75. It has no meaningful topic interpretation. Relabel the same categories as 10, 20, and 90, and the average becomes (180 + 280 + 720)/40 = 29.5, although no ticket has changed category.

The category proportions remain unchanged under either coding. This demonstrates why the measurement meaning, rather than the storage format, determines which calculations are meaningful. If tickets could receive multiple topics, the denominator and interpretation would need to be specified, and percentages across topics would not necessarily sum to 100%.

Illustrative case study

Fictional case study: Willow Support uses an AI assistant to classify incoming requests. It initially combines unclear messages with a category called other, making it difficult to distinguish genuine uncommon topics from classification uncertainty.

The team revises the codebook to preserve that distinction and samples labeled tickets for human review. Its dashboard shows category shares alongside the proportion of uncertain labels. This fictional case illustrates careful classification design, not a measured improvement in the assistant's accuracy or a claim that human labels are infallible.

Common misconceptions and mistakes

Qualitative does not mean low quality, subjective by definition, or impossible to analyze statistically. A clearly defined category can be measured consistently, while a numerical measurement can still be inaccurate.

Numerical category codes are another common source of confusion. Postal codes, identifiers, and nominal labels should not be averaged merely because software accepts the calculation. Conversely, counting category occurrences creates a valid quantitative summary. Finally, ordered responses do not automatically have equal spacing, and a language model's summary of interview material should not replace access to the original evidence and interpretation process.

Business use

Qualitative data help businesses understand what is happening and why people describe experiences in particular ways. They support segmentation, routing, product feedback, and evaluation of AI outputs.

The practical value depends on a stable vocabulary, clear definitions, and a record of how labels were assigned. Review categories when products or user behavior change, but version the classification so that trends remain interpretable. Avoid implying that a category name explains a cause. Where decisions depend on subtle meanings, combine structured summaries with representative source material and appropriate human review.

Check your understanding

A database changes topic codes from 1, 2, and 3 to 10, 20, and 90 without changing their meanings. Should topic proportions change?

Answer: No. The categories and counts are unchanged. A numerical mean of the codes may change, which illustrates why it is not an appropriate summary of nominal topic labels.

Related terms

Categorical variable
Ordinal variable
Quantitative data
Codebook
Proportion
Classification
nominal variable
annotation
frequency
qualitative research

Learn next

Compare nominal, ordinal, discrete, and continuous variables. Practice choosing summaries from the meaning of a variable, then study annotation quality and the interpretation of open-text material.

Sources

The Australian Bureau of Statistics defines qualitative data, categorical variables, and nominal and ordinal scales: https://www.abs.gov.au/statistics/understanding-statistics/statistical-terms-and-concepts/statistical-terms-and-concepts-glossary

The ticket arithmetic, coding comparison, and fictional case are original. The distinction between the glossary's categorical-data sense and broader qualitative research material is stated explicitly.

Previous: Qualitative comparative analysis Next: Qualitative physics