Statistics Glossary → Cross-loadings
Cross-loadingsconcept
Loadings that show how an indicator (item) relates to factors or constructs other than its own.
In factor analysis, each item should ideally load highly on one factor and weakly on the others (simple structure). An item that loads noticeably on two or more factors (e.g. above 0.30 or 0.40) is cross-loading; it may be measuring more than one construct and blurs the interpretation of the scale. Such items are usually reviewed or dropped. In confirmatory factor analysis cross-loadings are usually fixed at zero; if the data require them, fit suffers and the modification indices show it.
In PLS-SEM the cross-loading table is used for discriminant validity: each indicator should correlate more strongly with its own construct than with any other. This criterion is lenient and does not always catch problems; the HTMT ratio is a more sensitive alternative. In canonical correlation, a cross-loading is the correlation of each variable in one set with the canonical variate of the other set, and it shows which variables drive the relationship between the sets.
An everyday example
In a job-satisfaction survey, the item 'My salary reflects my effort' loads highly on the 'pay' factor as expected. The item 'I feel valued at work', however, loads about equally on the 'relations with managers' and the 'pay' factors; this cross-loading suggests the item measures two different things at once.
In MerQur
'Additional output: Cross-loadings' in Canonical Correlation and 'Extra output (cross-loadings)' in PLS-SEM; 'Additional output: Sort loadings by size, suppress small ones, cross-loading warning' in Exploratory Factor Analysis; in Confirmatory Factor Analysis a cross-loading can be added to the model through 'Additional model syntax'.
Analyses where this concept appears
Canonical CorrelationExploratory Factor Analysis (EFA)Confirmatory Factor Analysis (CFA)PLS-SEM (Partial Least Squares Path Modeling)