Description
Estimates construct reliability from optimally weighted CFA loadings, giving a better reliability index than Cronbach's alpha for UX scales with unequal item quality.
Cronbach's alpha is commonly misused as a measure of reliability, but it is actually a lower bound on reliability and is sensitive to the number of items. Coefficient H (Hancock & Mueller 2001) — also called maximal reliability — is theoretically superior: it represents the reliability of the construct estimated from a set of items using their optimal (CFA-derived) weights.
Attributes
- Discipline
- Survey & Psychometrics
- Loss type
- Extinct
- Stage
- Evaluation
- Data
- Quantitative
- Effort
- Days
Steps
- Compute Coefficient H for any UX scale:
- Run CFA on the scale items.
- (2) Extract standardized factor loadings λ_i.
- (3) Compute H = 1 / (1 + 1/Σ(λ_i²/(1-λ_i²))).
- (4) Compare H to alpha: if H substantially exceeds alpha, the scale has poor items that alpha is masking (some items are carrying all the reliability).
- (5) Target H > 0.80 for research purposes, > 0.90 for individual assessment. R: in semTools package: maximalRelia().
Sources
- Survey Scales — (various)
Also known as
- Maximal reliability
- Construct reliability H
Categories
- Analytics & Behaviour
- Survey & Psychometrics