Brander, Egan and Yeung, Estimating the Effects of Age on NHL player Performance, JQAS 2014..
I don't believe them or any other study I've seen. But if you want to use 22-29 for early arrivers and 24 -31 or so for later arrivals, that's about as reasonable as anything I've seen.
1) studies look at averages, but generally ignore the "early arrivers" and "late arrivers" effect. Early arrivers skew age curve younger, b/c they peak early, and thus are more likely to decline (simply because in a sample of top talents, injury, etc. will cause decay while it's hard for them to improve, it's like the guy who runs a sub-4.00 mile, easier to slow down that to get faster). Later arrivals peak later and then start declining from a later peak.
2) survivor issues, how do you handle players who haven't "declined" but got a serious injury? We're trying to measure decline due to age, not decline due to "events." Since injuries are generally stochastic (with some genetic component) you'd want to cull them from the data.
Though neither Hayes nor Voracek made this list:
Projecting the NHL's 15 Worst Contracts