I would do it under two circumstances: The first is either of two conditions: A random 15 case sample of a mixture of years had more than 16% bigger standard deviation than a random 15 case sample of same-year cases. Or, if 15 case random samples of two different years had averages more than about a grain and a half different from one another.
The second is if the two case years represented brass with different load histories. A case that's been reloaded more times than another will tend to have stiffer neck brass from the additional work hardening (until it is annealed again). I want consistent bullet pull to help keep start pressure the same.
Gunsite Orange Hat Family Member
CMP Certified GSM Master Instructor
NRA Certified Rifle Instructor
NRA Patron Member