APBRmetrics

Guy · Joined: 02 May 2007 Posts: 128

Baseball sabermetricians have long wrestled with the aging issue. You definitely have to look at the change in individual players, as you discovered, or survivor bias creates the illusion that players improve (or don't decline) with age.

However, the y-t-y change method also has a problem, which is regression to the mean. Players who play very well in year 1 will tend to decline in year 2. Players who play poorly in year 1 will tend to improve in year 2, BUT some of these players will not return in year 2 (or won't meet your playing time threshold) because of their poor play in year 1. So you capture all of the downward regression, but only some of the upward regression. This creates a bias that exaggerates decline. In baseball, hitters peak at about age 27, but the y-t-y delta method will tell you they start declining at age 22. You may have the same problem here.

A simple but pretty good approach is just measuring how many players have their best defensive season at age X. Alternatively, use your original method but first normalize defensive rating for each player, so each year's value compares the players' rating to his own career mean. However, that approach still leaves the problem that at the older ages, you are mainly looking at players whose skills declined at an unusually slow rate (otherwise, they wouldn't be playing).

Ryan J. Parker · Joined: 23 Mar 2007 Posts: 711 Location: Raleigh, NC

This aging stuff fascinates me, and it seems to be all the rage right now. I've been enjoying Phil Birnbaum's recent posts, like this one, that illustrate a lot of good points about quantifying aging with respect to baseball player statistics.
_________________
I am a basketball geek.