Understanding ANOVA requirements better.

Two ANOVA requirements are

1. Approximately Normal Data
2. Equal Variances

Online the normality requirement is often simply stated as the data should be approximately normal, but the real requirement is that the residuals generatd from the ANOVA test should be approximately normal. Assuming I am correct in this, here is my question.

When testing equal variances are we testing the groups of raw data or is it more accurate to say we should be testing different groups of the residuals created by the anova test?


Ambassador to the humans
When talking about the variances it really doesn't matter. If you break your data into the groups and then look at the variance of the groups you should end up with the same thing if you would have broken the residuals into groups and then looked at the variances of those.
Thank you for the quick response. Could you also possibly recommend a good stats books that covers concepts like checking the equality of the variance and normality for ANOVA in detail rather than the typical freshman level books that gloss of the topic saying "if the data is normal then your good". I have a B.S. in pure math but I have only taken basic undergraduate statistics. Again thanks for the help.