I have never seen any use of (or need for) anything resembling regression testing on the analog circuits I have worked with, but they have all been at the PWB or system level. When I did a web search, I found many references to regression testing of analog IC designs. Not having experience in that area, I can only guess that there might be more ways that sections of an analog IC can interfere with each other[1], and that this combines with the NRE needed to change an IC to make it so regression testing makes more sense than it does at the PWB level. Just a guess, of course, but *something* is driving the use of regression testing in analog IC design.
[1]: Most of the PWBs I have worked with make liberal use of decoupling caps and often have a ground plane, but I have noticed that with PWBs for toys -- which have few or no decoupling caps and long ground traces (single-sided PWBs) -- changing one section is more likely to mess up another section. Motorboating comes to mind. Toys are mostly digital with a few analog support sections, so I am thinking that a toy-style (cheap) design with a lot of different analog sections might call for regression testing of the analog circuits. It sure does help with software design...