Cronbach's alpha
Describe item covariance with explicit reliability assumptions
Coefficient alpha summarises the covariance structure of items contributing to a composite score. It equals composite reliability under suitable assumptions including essentially tau-equivalent items and uncorrelated errors. It can mislead when these fail, and a high value can result from many redundant items. Alpha does not test unidimensionality, demonstrate validity or establish stability over time.
Use as a clearly qualified item-based coefficient when the score construction and item assumptions are defensible, alongside examination of dimensionality and potentially more appropriate model-based reliability estimates.
Strengths
- Widely available and straightforward to reproduce
- Offers a familiar descriptive summary of item covariance
Limitations
- Sensitive to item count and correlated errors
- Unequal item loadings can make it an unsuitable reliability estimate
Know the boundary
An alpha above a conventional cut-off is not proof that a scale measures one construct well.