THE CALCULATION OF CONFIDENCE REGIONS FOR EIGENVECTORS
G. S. Watson · Australian Journal of Statistics · 1984
summary An explicit algorithm is given for constructing a confidence region on the appropriate unit sphere for an eigenvector, given a large sample. It is assumed the eigenvector corresponds to the largest eigenvalue of ExxT, a matrix with distinct eigenvalues, and that the estimation uses n‐1S̀xixixT. While the theory is generally applicable, the writer has in mind the special case where ∥x∥= 1 i.e. directional data.