Post-hoc anova on eigenvalues

[wagenrace]

"the f_classif test is really interesting. the patterns in the eigenvalues is also interesting. There seems to be a repeating pattern of a series of somewhat consecutive eigenvalues followed by a somewhat large gap. Without spending too much more time on the eigenface analysis (since you think it isn't looking like it will help all that much) is there a way to perform some sort of post-hoc test on the anova to determine which classes are significantly separable in each eigenvalue? It would be interesting if the classes are consistently different, but I suspect we'll see a few (or even only 1) consistently splitting out."

Originally posted by @gwaygenomics in #31