MerQuR+Statistics Glossary

Statistics Glossary → Factor (component) loadings

Factor (component) loadingsconcept

Coefficients showing how strongly an observed variable (item) is related to a factor or component; what the factors represent is interpreted from these loadings.

Factor analysis explains the observed items as linear combinations of a small number of latent factors plus an error term; each item's weight on each factor is its loading. After orthogonal rotation, a standardized loading is the correlation between the item and the factor, and its square shows how much of the item's variance that factor explains; the sum of an item's squared loadings across all factors is its communality. With oblique rotation the factors are correlated, so two matrices are distinguished: the pattern matrix gives each factor's unique contribution to the item, and the structure matrix gives the overall correlation between item and factor.

In practice, loadings above about 0.30-0.40 in absolute value are treated as meaningful, and each item is grouped with the factor on which it loads most highly. If an item loads similarly on several factors (a cross-loading), it is unclear which construct it measures, and the item may need revision. The sign of a loading means nothing on its own, since the direction of a factor is arbitrary, but signs within the same factor are read relative to one another. Component loadings in principal component analysis are read in a similar way, although components summarize the total variance without separating out an error term.

An everyday example

In a personality questionnaire, the items 'I feel comfortable at parties', 'I like meeting new people' and 'I draw energy from crowds' load 0.70-0.80 on the first factor, while 'I always plan my work' and 'I keep my things tidy' load highly on the second. The researcher labels the first factor 'extraversion' and the second 'orderliness'.

In MerQur

In Exploratory Factor Analysis, the 'Additional output: Sort loadings by size, suppress small ones, cross-loading warning' and 'Loading suppression cut' options; in Partial Least Squares Regression, 'Extra output: Weights/loadings/scores'. Principal Component Analysis, Confirmatory Factor Analysis, Canonical Correlation and PLS-SEM also report loading tables.

Analyses where this concept appears

Canonical CorrelationPrincipal Component Analysis (PCA)Cronbach's Alpha and Reliability AnalysisLikert Scale AnalysisExploratory Factor Analysis (EFA)Confirmatory Factor Analysis (CFA)Partial Least Squares Regression (PLS)PLS-SEM (Partial Least Squares Path Modeling)

Related concepts

Cross-loadingsFactor rotation (varimax, promax, oblimin)Communality