Average variance extracted
Assess indicator variance captured by a reflective construct
Average variance extracted summarises the share of reflective indicator variance accounted for by a construct relative to measurement error. It is usually used as evidence about convergent validity, not as a reliability coefficient. The Fornell-Larcker discriminant-validity comparison relates AVE to squared construct correlations, but can miss problems; additional evidence, including appropriate HTMT assessment, may be needed.
Use within a justified reflective measurement model to examine convergent evidence and one aspect of construct distinction, alongside loadings, model fit and theory-based validation.
Strengths
- Summarises indicator-level shared variance
- Makes weakly represented reflective constructs easier to identify
Limitations
- Thresholds can conceal heterogeneous item quality
- Fornell-Larcker comparisons may fail to detect construct overlap
Know the boundary
AVE is convergent-validity evidence, not a reliability coefficient or a general measure of model quality.