Stevens' scales (nominal/ordinal/interval/ratio)

Description

Classifies every measure by its scale type, nominal, ordinal, interval or ratio, to determine which transformations, summaries and statistical tests are legitimate.

Stevens defined four levels of measurement by the transformations that preserve their meaning: nominal (categories; any one-to-one relabelling), ordinal (order; any monotonic transformation), interval (equal differences; linear transformations) and ratio (true zero; multiplication by a constant). Each level permits different statistics: mode for nominal, median and percentiles for ordinal, mean and standard deviation for interval, and ratios and geometric means for ratio. UX regularly breaks these rules, for instance averaging single Likert items as if interval, or claiming that task time 'doubled' satisfaction. The debate about treating multi-item rating scales as interval is legitimate and well studied, but it should be a reasoned choice. Applying Stevens' scales is the first step of any measurement plan.

Attributes

Loss type
Vestigial
Effort
Hours

Steps

  1. List every variable in the study, from task success to satisfaction ratings to time.
  2. Classify each by scale type and record the justification (for example task time is ratio; a single 5-point item is ordinal; a validated multi-item scale such as SUS is commonly treated as interval).
  3. Specify permissible summaries and tests per variable before collecting data (for example median and Mann-Whitney for ordinal items).
  4. Check transformations and composite scores for legitimacy, for example not summing nominal codes.
  5. Where you treat ordinal data as interval, state the assumption and run a sensitivity analysis with an ordinal method.
  6. Review the reporting plan so claims match scale properties (no ratios on interval data).

Sources

  • Dix, A., Statistics for HCI: Making Sense of Quantitative Data (2020)
  • Sauro, J. & Lewis, J. R., Quantifying the User Experience (2nd ed., 2016)
  • Leavy, P., Research Design: Quantitative, Qualitative, Mixed Methods, Arts-Based, and Community-Based Participatory Research Approaches (2017)
  • Statistics for HCI (Alan Dix)
  • Quantifying the User Experience (Sauro & Lewis)
  • Research Design (Leavy)
  • The Inquiry Series, Ch 8 — Measurement Theory — Before You Measure Anything

Also known as

  • Levels of measurement
  • Scale types
  • Stevens' typology
  • Stevens' scales of measurement
  • Nominal, ordinal, interval, ratio
  • Variable types and measurement levels

Categories

  • Research Operations
  • Research Design & Rigor