Description
Tests whether items form a cumulative hierarchy, so that endorsing a harder item implies endorsing all easier ones, revealing ordered stages of adoption or capability.
A Guttman scale is cumulative: if a respondent agrees with item 4, they also agree with items 1-3. The coefficient of reproducibility (target above 0.9) measures how well responses fit this pattern. Mokken scale analysis provides a probabilistic, nonparametric version with Loevinger's H coefficients (H above 0.3 weak, above 0.5 strong scale). Cumulative scales are natural for capability or adoption ladders: digital skills, feature maturity, privacy protective behaviours, automation trust. They reveal the order in which people acquire behaviours, which directly informs progressive onboarding.
Attributes
- Discipline
- Survey & Psychometrics
- Loss type
- Extinct
- Stage
- Definition
- Data
- Quantitative
- Effort
- Days
Steps
- Draft items expected to form a cumulative hierarchy (e.g. increasingly advanced feature uses).
- Collect binary or polytomous responses from a sample of users.
- Order items by popularity and respondents by total score; inspect the response matrix for errors.
- Compute the coefficient of reproducibility, or run Mokken scale analysis (H and Hi coefficients, monotonicity checks).
- Remove or revise items that break the hierarchy, and test invariant item ordering across groups.
- Use the scale to place users on stages and design progression accordingly.
Sources
- Guttman, A basis for scaling qualitative data (1944), American Sociological Review
- Mokken, A Theory and Procedure of Scale Analysis (1971)
- van der Ark, Mokken scale analysis in R (2007), Journal of Statistical Software
Also known as
- Scalogram analysis
- Cumulative scaling
- Mokken scaling (probabilistic version)
- Bogardus social distance scale
Categories
- Analytics & Behaviour
- Survey & Psychometrics