Ordinal vs. interval distinction

Description

Determines whether rating data can be treated as interval (equal distances) or only ordinal (ranked), and chooses summaries and models that respect the scale's properties.

Stevens defined four levels of measurement: nominal, ordinal, interval and ratio, each permitting different operations. A single Likert item, for example 1 to 5 agreement, is ordinal: the distance from 'agree' to 'strongly agree' is not guaranteed to equal that from 'neutral' to 'agree'. Means and t-tests assume interval properties. Norman (2010) argued parametric tests are robust in many cases, especially for multi-item scale totals, while Liddell and Kruschke (2018) showed that treating ordinal data as metric can produce wrong conclusions, including inverted effects, when distributions differ in shape. The practical rule is to treat single items as ordinal, consider summed multi-item scales as approximately interval if validated, and use ordinal models when distributions are skewed, have ceiling effects or differ in variance.

Attributes

Loss type
Vestigial
Effort
Hours

Steps

  1. Classify each measure: single rating item, multi-item validated scale (for example SUS or UMUX-Lite) or a behavioural count or time.
  2. For single items, report the distribution (percentages per response option) and medians, and use top-box or top-two-box shares where interpretable.
  3. Check distributions for ceiling or floor effects, skew and unequal variances between groups.
  4. Analyse single items with ordinal models (cumulative logit or probit regression) or rank-based tests, and reserve means and t-tests for validated multi-item scales with approximately symmetric distributions.
  5. Where both approaches are plausible, run both and report whether conclusions differ.
  6. Communicate results in terms of response distributions or probabilities (for example 'the probability of rating 4 or 5 rose from 52% to 61%').

Sources

  • Statistics for Behavioral Sciences
  • Discovering Statistics via SPSS (Field)
  • Survey Methodology (Groves)
  • The Inquiry Series, Ch 42 — Ordinal Logic — Treating Likert & Survey Data Correctly

Also known as

  • Levels of measurement
  • Scale type analysis
  • Treating Likert data as ordinal

Categories

  • Analytics & Behaviour
  • Survey & Psychometrics