Anova followed by paired t-tets for alpha diversity metrics

Dear Qiime2 developers!
I performed Anova test on alpha diversity metrics, followed by pairwise t-tests to test differences between treatments for three sample-types, separately for each sample-type. As a result, I got some significant differences for one of the treatments in one sample-type. As far as I know, t-test is a parametric test, and should be applied to normally distributed data. So I checked it with Shapiro-Wilk test, feeding alpha diversity metrics for each of sample-types, and found that alpha diversity metrics are not normally distributed for a sample-type, in which significant difference was observed, meanwhile in other samples metrics are normally distributed.
So isn’t it a good idea to add Shapiro-Wilk test result to the output of anova plugin to indicate data distribution? Or anova plugin performs some kind of data transformation under the hood?

Hi @timanix !

Correct — ANOVA and t-tests assume that data are normally distributed. It is up to users to confirm that the assumptions for these and other tests are being met.

Sure, I like the idea of adding this to the anova visualizer… feel free to open an issue and contribute if you are interested :wink:

1 Like

Hi @Nicholas_Bokulich
Thank you for clarifications! I will open an issue

1 Like