Eigenvalue, parallel analysis, loading and rotation

The eigenvalue says how much of the differences in all questions a component captures. The Kaiser criterion keeps components with an eigenvalue above 1; it often gives too many. Parallel analysis compares the eigenvalues with those from random data of the same size and keeps only what lies above them - it is the more reliable suggestion. Rotation turns the solution so that each question belongs as clearly as possible to one component: varimax keeps the components independent, promax lets the factors be correlated.

Example: Learning at university, 6 questions: 2 components, 63% captured together. Question 7 loads .89 on component 1 and nothing on component 2.

Caution: Loadings below .30 are hidden. If a question loads above .30 on two components, it does not clearly belong to one.