1

I have data for several samples, each sample having a different number of measurements done on the same physical magnitude. Some samples are "good" meaning their variance is low, and other are "bad" (high variances observed, relative to the "good"). I need to flag the "bad" ones on the basis of their higher variance. I learned I could use a Brown-Forsythe or Levene test for unequal variances. If the variances among groups are shown to differ by this test, I need to further know the significance of the pairwise differences. What test should I use for this? (It would be the equivalent of an ANOVA with post-hoc comparisons, but looking at the variances, not the means) Many thanks in advance.

user93631
  • 11
  • 1
  • 1
    See this link for an explanation on how to perform post-hoc comparisons of variances after performing a Levene test in R: https://stackoverflow.com/questions/43646987/multiple-comparison-post-hoc-test-for-levenes-test. – Isabella Ghement May 26 '18 at 16:41
  • 1
    Thank you Isabella. I also found [https://stats.stackexchange.com/questions/337335/post-hoc-test-to-determine-difference-in-variance]. I am looking for a confirmation as to whether this approach has any objections. For instance, as the residuals get folded by applying absolute value to them, their distribution would be half-normal and I don't know how this affects the regular post-hoc tests. Any comment? – user93631 May 29 '18 at 13:48

0 Answers0