1

My PCA with prcomp in R results in very low "loadings" (i.e. eigenvectors, see figure below).

I've tried a rotation with rotated_PCA = varimax(PCA_result$rotation) like suggested here, but it didn't help. Does that mean my scores don't relate to the original variables? The following PC loadings are low, too. I couldn't find any explanation in literature for interpreting low loadings. And it is still the same PC analysis like this one.

loadings Plot over time

sequoia
  • 143
  • That seems very odd. This would seem to indicate that there is very little variance being captured by the PC. Be careful not to mix up 'loading' with 'eigenvector' - the two are different, as described here: https://stats.stackexchange.com/questions/143905/loadings-vs-eigenvectors-in-pca-when-to-use-one-or-another – Don Walpola Aug 31 '18 at 17:34
  • @DonWalpola that's why I put loading in quotation mark, prcomp returns eigenvectors. And the first PC represents 56% of original variance or do I misunderstand your comment? – sequoia Sep 01 '18 at 07:49
  • So the loadings would then represent, in a sense, the proportional contribution each of the original covariates contributes to the first PC - you must be in a high dimensional space, with numerous independent variables. The sum of the squares of all the loadings equals the eigenvalue associated with the eigenvector/ first PC. Having low loadings on each of your original variables just means each original variable only makes a small contribution to the PC, i.e., no original variable points too much in the direction of the PC. – Don Walpola Sep 02 '18 at 02:20

0 Answers0