Average Variance Extracted (AVE)
Statistics & analysis
The average share of variance a latent construct captures from its indicators; AVE of 0.50 or higher supports convergent validity.
AVE is the average proportion of variance a latent construct explains in its own indicators, computed as the mean of the squared standardised loadings. It is the standard measure of convergent validity in both CFA and PLS-SEM.
The accepted threshold is 0.50. At or above it, the construct explains more of its indicators' variance than measurement error does; below it, the items carry more error than signal. The usual remedies are dropping weakly loading indicators or reconsidering whether the items genuinely tap a single construct.
AVE also underpins the Fornell-Larcker criterion for discriminant validity, which requires the square root of a construct's AVE to exceed its correlation with every other construct. Because Fornell-Larcker is now regarded as less sensitive than the HTMT ratio, many reviewers expect to see both reported.
Where it's used
- Assessing convergent validity of measurement models in SEM
Software used
Related concepts
Guides on this topic
- Convergent and discriminant validity thresholds (AVE, HTMT, Fornell-Larcker)
- Composite reliability explained
- HTMT ratio for discriminant validity
- SPSS vs SmartPLS: which should you use?
- Model fit indices in AMOS explained
- Interpreting SmartPLS bootstrapping results
- What is SmartPLS? PLS-SEM explained
- What is jamovi? The free SPSS alternative
Guides that use this concept
All guidesNot sure where to start?
Book a free 15-minute consult. We'll map your next three steps — no obligation.