Statistics Glossary → KMO measure of sampling adequacy and Bartlett's test of sphericity
KMO measure of sampling adequacy and Bartlett's test of sphericityconcept
Two preliminary checks of whether a correlation matrix is suitable for factor analysis: the Kaiser-Meyer-Olkin measure and Bartlett's test of sphericity.
KMO compares the simple correlations between variables with their partial correlations: KMO = Σr² / (Σr² + Σa²), where a is the partial correlation with all other variables held constant. If the variables share common factors, the partial correlations stay small and KMO approaches 1. By Kaiser's labels, values above 0.90 are marvellous, in the 0.80s meritorious, the 0.70s middling, the 0.60s mediocre, the 0.50s miserable and below 0.50 unacceptable. The MSA values computed for each variable (the diagonal of the anti-image matrix) point to problem items.
Bartlett's test checks the null hypothesis that the correlation matrix is an identity matrix (that the variables are uncorrelated): χ² = −(n − 1 − (2p + 5)/6) · ln|R|, with df = p(p − 1)/2. A significant result shows that factor analysis is not pointless, but because it is almost always significant in large samples it is not proof of adequacy on its own. Neither measure shows whether a factor structure exists or how many factors there are; they only say whether the analysis is worth trying.
An everyday example
A researcher gives a 15-item attitude scale to 300 people. KMO = 0.84 and Bartlett's χ² is significant, so the matrix is suitable for factor analysis. One item, however, has an MSA of 0.41; it shares little with the others and dropping it from the scale is considered.
In MerQur
In Principal Component Analysis and Exploratory Factor Analysis, KMO and Bartlett's test are reported together; per-variable MSA values are given by 'Communalities (h²) + variable MSA' in Principal Component Analysis and with the anti-image matrices in Exploratory Factor Analysis.
Analyses where this concept appears
Principal Component Analysis (PCA)Exploratory Factor Analysis (EFA)
Related concepts
Factor extraction method (ML, PAF, minres)Covariance and correlation matrix