Measurement scales and data types
A variable can name groups, order them or measure amounts; the stored numbers do not decide what comparisons mean.
Fit
When it matters, and when it does not
It matters when
- You define an outcome, code a dataset or plan descriptive summaries.
- You need to distinguish a measured amount from a category given numeric codes.
It matters less when
- You already know the data type and need an effect estimate. Start with the estimand and the study structure. Read: The estimand
Before you start
What to check in your study
- What each value meansRecord units, category labels, ordering and missing-value codes.
- The outcome and contrast you wantoutcomeTypesummaryMeasureA mean difference, risk difference and shift across ordered categories ask different questions.
- How observations are relatedclusteredrepeatedMeasuresIdentify pairs, repeated measures and clusters before choosing an analysis.
- Equal numeric gaps need a measurement justification.How to check: Ordered labels alone do not establish equal distances between categories.
- Codes retain the intended meaning.How to check: Keep unknown values separate from genuine absence. Changing a code must not change the clinical interpretation.
Plain language
How it works
Nominal values name categories with no order, such as blood group. Binary values have two categories. Ordinal values have an order without assured equal gaps. Interval scales have equal units but no meaningful zero, such as Celsius temperature. Ratio scales also have a meaningful zero, such as mass. These distinctions concern meaning, not storage format.
Counts are discrete amounts; time to an event also needs a time origin and censoring rules. A comorbidity index is a scored quantity, not just a disease yes/no flag. Its numeric summary does not prove equal clinical gaps. Stevens describes the scale distinctions; Charlson and colleagues developed and validated a weighted comorbidity index.
Worked example
A worked example
- Setting
- An invented clinic records pain categories and a separate improvement outcome in 80 patients.
- Question
- What proportion reports improvement?
- Approach
- Keep mild, moderate and severe as ordered categories. Code improvement as yes or no; 12 patients improve.
- Result
- 12 of 80 patients report improvement, 15 in 100.
- What it means
- Improvement is binary, while pain severity is ordinal. Numeric category codes would not make pain an equal-interval measurement.
In practice
Recent studies PubMed search · not vetted
Watch out
Common mistakes
The one to remember
A numeric code does not make a category continuous. Measurement scale alone cannot choose an analysis: the estimand, pairing, clustering and assumptions still govern.
- Choosing a test from the scale label alone.Instead: Check the estimand, distribution, dependence and model assumptions. No scale-to-test table resolves all of them.
- Treating a coded category as a measured amount.Instead: Inspect labels and units. A numeric identifier remains nominal; a single ordered rating remains ordinal.
- Assuming every score has equal units or a meaningful zero.Instead: Check how the score was built and validated. A multi-item score needs evidence for its intended interpretation.
Inside Mallard
How Mallard uses this
Mallard types each Table 1 variable from its name and treats a comorbidity index as continuous.
concept.measurement_scales, the id a plan cue records.Related Learn pages
Related designs and methods
Evidence
Sources & further reading
- On the Theory of Scales of Measurement (opens in a new tab) · Science · doi 10.1126/science.103.2684.677
- A new method of classifying prognostic comorbidity in longitudinal studies: development and validation (opens in a new tab) · J Chronic Dis · doi 10.1016/0021-9681(87)90171-8