Composite reliability
Estimate reliability from a specified measurement model
Composite reliability estimates the proportion of a score's variance attributable to its intended common factors under a measurement model. Congeneric models allow unequal loadings, unlike the assumptions needed for alpha to equal reliability. The calculation depends on the score weights, factor structure and error specification. It should not be confused with AVE, which concerns extracted variance at the indicator level.
Use for a reflective multi-item score when an identified measurement model fits substantively and statistically and unequal indicator loadings make a congeneric approach appropriate.
Strengths
- Accommodates unequal item loadings
- Can make measurement-error assumptions explicit
Limitations
- Misspecified factors or residual correlations distort estimates
- Different formulas and scoring targets can produce different coefficients
Know the boundary
A high composite reliability does not establish discriminant validity or justify a misspecified factor model.