Statistics Glossary → Scree plot
Scree plotconcept
A plot of the eigenvalues from a principal component or factor analysis in decreasing order, used to help decide how many components or factors to keep.
The component number is on the horizontal axis and its eigenvalue (the variance it explains) on the vertical axis. Following Cattell (1966), the curve first drops steeply and then flattens out from an 'elbow'. The components before the elbow represent meaningful structure, and the flat tail represents random noise. The name comes from its resemblance to the rubble (scree) that collects at the foot of a mountain.
The location of the elbow is often ambiguous, different readers may choose different numbers, and there may be more than one elbow. The scree plot should therefore be used not alone but together with parallel analysis, the MAP test, the eigenvalue > 1 (Kaiser) rule and the proportion of total variance explained. Simulation studies have found parallel analysis to be more reliable, while the Kaiser rule often suggests too many factors. The theoretical interpretability of the components also matters in the final decision.
An everyday example
A questionnaire has 20 items. In the scree plot the first three eigenvalues are 6.1, 3.4 and 2.2, and from the fourth onward the values decline slowly: 0.9, 0.8, 0.7. Because the elbow comes after the third component, the plot suggests that the questionnaire measures three main dimensions.
In MerQur
In principal component analysis it is drawn by choosing 'Scree' under the 'Chart' option. In multidimensional scaling, 'Stress scree (1..k dimensions)' computes Stress-1 for each number of dimensions from 1 to k, and it is drawn by choosing 'Stress scree' under the 'Chart' option.
Analyses where this concept appears
Principal Component Analysis (PCA)Multidimensional Scaling (MDS)
Related concepts
Eigenvalue and eigenvectorParallel analysis (Horn) and the MAP criterion