3
$\begingroup$

In statistics, given a covariance matrix $\Sigma$ with singular values $\sigma_1 \ge \sigma_2 \ge \ldots \ge \sigma_p$, is the ratio of its spectral norm to the Frobenius norm, i.e the ratio $\dfrac{\sigma_1}{\sqrt{\sum_{i=1}^p \sigma_j^2}}$ of any interest ? That is, does this ratio appear naturally in the analysis of certain procedures, algorithms, convergence limits, etc. ?

Thanks in advance!

$\endgroup$

1 Answer 1

2
$\begingroup$

It could be taken as an indicator of how close the matrix is to a rank 1 matrix. In MANOVA you have cases where this matters.

$\endgroup$
3
  • $\begingroup$ OK, thanks for mentioning MANOVA (never heard of that!). Any explicity reference of the how this ratio makes its way into MANOVA ? About low ranksity, more generally, it's not hard to prove that $1 / rank(\Sigma) \le \sigma_1 / \|\Sigma\|_F \le 1$. For example, see math.stackexchange.com/a/3638563/168758. $\endgroup$
    – dohmatob
    Apr 22, 2020 at 17:40
  • $\begingroup$ In 1-way MANOVA one wants to test equality of means when the response is multivariate. The statistics are functions of the eigenvalues of $E^{-1}H$ where $H = n\sum_i(Y_{i.} - Y_{..})(Y_{i.} - Y_{..})^T$. Now, if the centroids of the groups are nearly aligned, a test statistic such as the Roy's maximum root will have better power than alternative statistics, such as Pillai's or Lawley-Hotelling. Too long to explain in detail in a comment. $\endgroup$
    – F. Tusell
    Apr 22, 2020 at 18:01
  • $\begingroup$ OK, great. Thanks! BTW, the comment could be made part of the answer. $\endgroup$
    – dohmatob
    Apr 22, 2020 at 18:19

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.