13.9 Things to watch out for

  • Eigenvalues are estimated variances of the PC’s and so are subject to large sample variations.
  • The size of variance of last few principal components can be useful as indicator of multicollinearity among original variables
  • Principal components derived from standardized variables differ from those derived from original variables
  • Important that measurements are accurate, especially for detection of collinearity
Arbitrary cutoff points should not be taken too seriously.