I'm almost done with the retrospective on Established Win Shares Levels (EWSL, explained here), before I kick into the preseason previews, and in so doing, I'm looking again at the age adjustments. (I've looked at these previously here, here after 2004, and here after 2005, and here after 2006).

The great thing about doing something like EWSL as an ongoing project is that the data becomes progressively more stable over time: I now have four years of results to work from in evaluating how players tend to perform at each age relative to their adjusted Established Win Shares Levels, and thus can have progressively more confidence in the age adjustments I use going forward. For example, the more years of data I have, the less influenced it will be by a single generation of exceptional players born in a particular year.

Let's start with the 4-year results for the non-pitchers:

Non-Pitchers 2004-07:

As I have noted in previous reviews, the rapid rise of young players and their gradual fall from age 29 on is a powerful pattern in the data, and one that grows smoother with each year's additional data - on average, players lose more than 10% of their established value each year from age 30-34, and more than 25% each year from 35-38, and nearly half every year from 40 on - and that's just the people who hang onto their jobs. After age 32, the number of players holding jobs at all entering a season really starts to drop off. As I've explained before, the nature of any established performance level will exaggerate the upward and downward trajectory of player aging, since a 25-year-old is still being partly compared to his 22-year-old self, while a 35-year-old is still being partly compared to his 32-year-old self - but the pattern as a whole is still unmistakable.

Here's this year's data on its own:

Non-Pitchers (2007):

Rob McMillin notes a "Lake Woebegone effect" in this year's data, in that EWSL underestimated a lot of teams' performances overall - that's a bit of a misunderstanding of the team data, as actually more teams' 23-man rosters under-performed than over-performed, the difference being made up by guys who were not in my EWSL calculations before the season. But there's a larger point here: in fact, 2007 seems to have been an unusually tough year for non-pitchers trying to match their established performance levels. You can speculate why by looking at the age distributions or the possibility that players were being weaned off performance enhancing drugs, but the effect is there in the data - while small sample sizes and the other issues with very young players may be the culprits for less dramatic improvements by the age 22-23 crowd, the declines began at age 27 this season rather than the traditional 29, and the 30-year-olds in particular had a horrendous year, with large declines in particular by Ben Broussard, Travis Hafner, Lyle Overbay, Robb Quinlan, Craig Wilson, Michael Barrett, AJ Pierzynski, Jay Gibbons, Reed Johnson, Andruw Jones, Craig Monroe, and Adam Everett all playing a role. The 34-year-olds also had a rough year, headed by Jay Payton, Cliff Floyd, Brady Clark, Vance Wilson, and Chris Coste, but this looks more like a small-sample-size issue than anything.

Now, the pitchers:

Pitchers (2004-07):

In general, the rule still holds that the pitchers as a group start to fall off earlier than the hitters. The age 21 and under cohort continues to be dominated by the career of Felix Hernandez, so it's not terribly predictive - of course, guys who open a season in a rotation at that age tend to be pretty much unique anyway. By contrast, you can see if you follow these things from year to year that the performance of the over-40 crowd has been declining for a couple of years, mainly because it's the same handful of guys who have pitched well into their 40s now and are reaching the end of the line. That will continue in 2008, as there were no 39-year-old pitchers in the sample in 2007. In general, you can see that there are a lot of young (age 23-26 in 2007) pitchers in the game today, with a huge dropoff between 31 and 33. As a whole the results for the late 30s are still influenced by a small sample size. The 2007 data:

Pitchers (2007):

The 23 and 24 year olds had a bad year as a group, the kind of bad year you would not see for a group of hitters that age. Among the 23 year olds, the Marlins were almost wholly responsible (Josh Johnson, Scott Olsen, Anibal Sanchez). Zach Duke and Ervin Santana were the worst offenders in the age 24 group, with the rest being mainly low-level pitchers zeroing out. The 34 year olds had a horrible year for the same reason 34 year old pitchers usually do - Bartolo Colon, Jason Schmidt, Armando Benitez. The 36 year olds had a big year just due to a small sample size - Al Reyes, Paul Byrd and Miguel Batista were the big movers.