Regression

Topic of regression testing came up. I take it to mean testing software against "regressions". That is, creating automated tests for every new feature or bug fix. These tests are then run against any future change to make sure the feature continues to work and that bugs once fixed never occur again. Having a previously working feature break or a bug re-occur is called a "regression" -- thus the name of the test. I believe I am supported in my definition, for example by Wikipedia and Webopedia.

I was in a recent discussion where it was taken to mean more than that. Proper hierarchical ordering of tests to ensure only the necessary tests are run was mentioned. I'm familiar with selecting only affected tests. But it sounds like there's more to it than this. I need to learn more here.

I mentioned it to my wife and she showed me Super Crunchers (Ian Ayres, Bantam, 2007). Ayres mentions regressions (defined pg. 23). He uses it to mean formulas derived from statistical data that predict future results. I learned something! I tend to call this "correlation" because the mathematics behind Ayres regression formulas is what I call correlation processing. I've used it in a signal processing context, but I believe it is similar to what Ayres is talking about. I never knew before that when statisticians referred to "regressions" this is what they were referring to.

Is there software testing, say for performance, that uses correlation processing to produce a "regression" that explains what is happening? Like which factors most contribute to performance hits? Or am I just being silly? Need to learn more.