Statistical & Clinical Analysis
McDonald’s Omega (Composite Reliability) & Fornell-Larcker AVE Formula in CFA
Replace tau-equivalent Cronbach’s alpha assumptions in Confirmatory Factor Analysis (CFA / PLS-SEM) using Roderick McDonald’s congeneric omega = (SUM(lambda))² / ((SUM(lambda))² + SUM(1 - lambda²)) and Claes Fornell & David Larcker’s Average Variance Extracted AVE = SUMPRODUCT(lambda, lambda) / k.
Why modern psychometrics & SEM journals require McDonald’s Omega (>= 0.70) and Fornell-Larcker AVE (>= 0.50)!
Cronbach’s alpha assumes strict tau-equivalence (every survey item loads identically on the latent factor), systematically underestimating scale reliability when standardized CFA factor loadings lambda_i differ! Roderick P. McDonald (1999, Test Theory: A Unified Treatment) and Claes Fornell & David F. Larcker (1981, Journal of Marketing Research) derived congeneric Composite Reliability (omega) directly from standardized single-factor loadings lambda_1..lambda_k (where each item’s unique error variance is theta_i = 1 - lambda_i^2): omega = (SUM(lambda))^2 / ((SUM(lambda))^2 + (k - SUM(lambda^2)))!
Alongside omega >= 0.70 internal consistency, Fornell and Larcker defined convergent validity via Average Variance Extracted AVE = SUM(lambda^2) / k >= 0.50 (meaning the latent construct explains > 50% of item variance, and SQRT(AVE) must exceed inter-construct correlations for discriminant validity)! For 4 items with loadings [0.82, 0.78, 0.85, 0.75], omega = 0.8771 and AVE = 0.6415—whereas weak loadings [0.55, 0.60, 0.58, 0.52] fail convergent validity at AVE = 0.3173 (< 0.50)!
| 4-Item CFA Standardized Loadings | McDonald’s Omega / CR (Cutoff >= 0.70) | Fornell-Larcker AVE (Cutoff >= 0.50) | Convergent Validity Verdict |
|---|---|---|---|
| [0.82, 0.78, 0.85, 0.75] | 0.8771 | 0.6415 (64.15% shared variance) | Strong Reliability & Convergent Validity |
| [0.55, 0.60, 0.58, 0.52] | 0.6502 (< 0.70) | 0.3173 (< 0.50 — Error dominates!) | Inadequate Construct Validity |
| Heywood Case |lambda_i| >= 1.0 | Negative Unique Variance! | "Invalid Standardized Loadings (Requires 0 < lambda < 1)" | CFA Identification Guard |
McDonald’s Congeneric Omega / Jöreskog Composite Reliability Formula
Evaluates (SUM(B2:B5))^2 / ((SUM(B2:B5))^2 + (COUNT(B2:B5) - SUMPRODUCT(B2:B5, B2:B5))) across standardized factor loadings in (0, 1).
=IF(OR(MIN(B2:B5) <= 0, MAX(B2:B5) >= 1), "Invalid Standardized Loadings (Requires 0 < lambda < 1)", ROUND((SUM(B2:B5) ^ 2) / ((SUM(B2:B5) ^ 2) + (COUNT(B2:B5) - SUMPRODUCT(B2:B5, B2:B5))), 4))Fornell-Larcker Average Variance Extracted AVE = SUM(lambda²) / k
Computes mean squared standardized loading across the k indicators (>= 0.50 confirms convergent validity).
=IF(OR(MIN(B2:B5) <= 0, MAX(B2:B5) >= 1), "Invalid Standardized Loadings (Requires 0 < lambda < 1)", ROUND(SUMPRODUCT(B2:B5, B2:B5) / COUNT(B2:B5), 4))Try the example
Use the original values or change them to test your understanding.