Summarize an Ordinal Scale Without Hiding Its Categories

Establish what the order means

An ordinal outcome has categories with a meaningful order, such as none, mild, moderate, and severe. The distances between adjacent categories are not automatically equal. Write the labels and their direction before choosing a summary: does a larger code indicate greater severity, better function, or something else?

Cochrane's discussion of ordinal outcomes and measurement scales distinguishes ordered categories and discusses possible analysis approaches. The appropriate treatment depends on the scale and question; a number stored in a spreadsheet does not by itself make a variable continuous.

Preserve the distribution

Begin with counts and percentages for every category, in the intended order. Show the number with an observed response and the number without one. Specify whether percentages use all eligible participants or only those with a recorded response, and keep that denominator consistent across the table.

Do not alphabetize response labels if doing so destroys their order. Keep zero-count categories visible when they belong to the scale. If different sites use different coding schemes, map codes to verified labels before combining results; code 1 may represent the opposite end of the scale in another export.

Work through a fictional distribution

Suppose 40 respondents report none, mild, moderate, or severe symptoms, with counts 8, 12, 15, and 5. The percentages are 20%, 30%, 37.5%, and 12.5%. The cumulative count reaches 20 at mild and 35 at moderate, so the two central observations lie in different categories.

For this even-sized sample, a numerical median based on codes 1–4 would be 2.5. That value is not an actual response category and should not be presented as though participants selected it. Showing the full category distribution avoids concealing this feature. These data are invented solely for illustration.

Keep scoring decisions explicit

Distinguish a single ordinal item from a validated composite score constructed from several items. Follow the instrument's authorized scoring instructions for reversed items, missing responses, and totals. Do not create a new average score simply because the items share the same response options.

If categories are combined, record the scientific rationale and prespecified rule. Collapsing categories after seeing a preferred p-value changes the analysis and can discard useful information. Similarly, treating a scale as continuous requires justification rather than a universal rule that every ordinal measure must or must not be averaged.

End with the labels, scoring reference, denominator, category distribution, and any transformation in the analysis record. If planning an ordinal regression or another inferential model, assess that model's assumptions separately. A clear descriptive table is the starting point for interpretation, not proof that a particular model is suitable.

Source checked on 1 November 2026; fictional counts independently recalculated.