Description
Models the probability of item responses as a function of a latent trait and item properties, enabling precise scale calibration, item selection and adaptive questionnaires.
Item response theory models how a person's level on a latent trait, such as perceived usability, digital literacy or trust, and the properties of each item jointly determine response probabilities. Item parameters include difficulty (location) and discrimination, and polytomous models such as the graded response model handle Likert items. Unlike classical test theory, IRT gives item-level information, shows where on the trait each item measures precisely, supports equating across versions, and makes computerised adaptive testing possible. For UX scales, IRT can identify redundant items, create short forms that preserve precision and check that items function equally across groups through differential item functioning analysis. Rigour comes from adequate samples (often 300 or more), checking unidimensionality and local independence, model fit assessment and DIF testing.
Attributes
- Discipline
- Survey & Psychometrics
- Loss type
- Latent
- Stage
- Evaluation
- Data
- Quantitative
- Effort
- Weeks
Steps
- Define the latent trait and assemble a pool of candidate items.
- Collect responses from a large, diverse sample, typically 300 or more.
- Check unidimensionality (for example with factor analysis) and local independence.
- Fit an appropriate model, such as Rasch, 2PL or graded response, and assess item and model fit.
- Inspect item parameters and information curves to find redundant or poorly discriminating items.
- Test for differential item functioning across key groups such as language or age.
- Build short forms or adaptive versions and validate them against the full scale.
Sources
- Quantifying the User Experience
- Item Response Theory for Psychologists (Embretson & Reise)
Also known as
- IRT
- Latent trait theory
- Rasch model
Categories
- Analytics & Behaviour
- Survey & Psychometrics