Description
Build an interval scale on a set of items from judgements of which of two differences is larger, after checking that judged differences add up along one continuum.
Many UX quantities cannot be physically concatenated, yet researchers want equal-interval scales for them. Difference measurement obtains equal spacing from an ordering of intervals: which of two pairs differs more, bisection settings, rating-scale distances, or differences inferred from preferences. The ordering yields an interval scale if it is a weak order, reverses sign when pairs are reversed, and satisfies weak monotonicity: if a-b matches a'-b' and b-c matches b'-c', then a-c must match a'-c'. Solvability and Archimedean conditions complete the theory. A standard sequence of equally spaced items can then be built and counted; when exact matches cannot be observed, strict inequalities give approximate sequences or bounds on scale values. Failure of weak monotonicity signals that judged differences are not one-dimensional.
Attributes
- Discipline
- Survey & Psychometrics
- Loss type
- Unmigrated
- Stage
- Discovery, Evaluation
- Data
- Quantitative
- Effort
- Days
Steps
- Collect an ordering of item pairs by direct comparison of differences, bisection (choose the item halfway between two others), rating positions or inference from preferences.
- Check that the ordering of intervals is transitive and that reversing a pair reverses the judgement.
- Test weak monotonicity: when adjacent intervals match, check that the combined intervals also match, and count violations.
- Build a standard sequence of equally spaced items and check that interval ratios do not depend on which sequence is used.
- Where exact matches are unavailable, use clear strict inequalities to form approximate sequences or to bound scale values by solving the inequality system.
- Report scale values with the violations found and, for cross-modality data, check the cross-modality form of monotonicity before claiming power-related ratio scales.
Sources
- Foundations of Measurement, Vol. 1: Additive and Polynomial Representations (Krantz et al., 1971)
Also known as
- Algebraic-difference structure
- Positive-difference structure
- Ordering of intervals
- Bisection scaling
- Weak monotonicity test
- Approximate standard sequences
Categories
- Analytics & Behaviour
- Survey & Psychometrics