December 3, 2003
Can Of Corn
Last time, we looked at cumulative run differentials as a way of evaluating an organization's farm system. We're going to revisit that idea, but this time we'll attempt to adjust for age. Organizations, natch, have different drafting strategies and promotion philosophies, which leads to some age variance from level to level. Age relative to peer group is a vital analytical component when scrutinizing individual prospects, and it should also be a factor on the systemic level. And so it shall be.
Another change this time around is that I've narrowed the focus to each organization's full-season affiliates (Triple-A, Double-A, High-A and Low-A). I made this decision because you see quite a bit of variation in how teams flesh out the lower rungs of their systems. For instance, in 2003 nine teams opted to field multiple rookie-level teams and no short-season affiliate at all. This makes system-wide comparisons at the lower levels a bit nettlesome and misleading. I'd also suggest that it's appropriate to place the emphasis on those levels closest to the major leagues.
Here are the average ages for each organization's full-season affiliates, ranked youngest to oldest:
And here are the unadjusted cumulative run differentials for each organization's four full-season affiliates:
The problem with adjusting for age is that the low variance among ages (standard deviation = 0.43) relative to run differentials (standard deviation = 188.65) means that the importance of age is likely to be severely downplayed in any kind of quick-and-dirty factoring setup. By that I mean calculating a "park factor" proxy using ages and refiguring runs that way usually yields an adjustment of around 1.5 runs; that's plainly undervaluing the effect that age differences should have on the final figure.
So I've decided to go with an approach that places more weight on average ages. Here's how it goes: Organizations will receive scores based on how many standard deviations they were better or worse they were than the mean run differential and mean age of all 120 full-season affiliates. (Incidentally, this methodology owes at least a little something to Rob Neyer's and Eddie Epstein's excellent book from a few years ago. Furthermore, the scores will be weighted by a factor of 6.7 for the run differential standard deviation score and 3.3 for the age standard deviation score. Then the sum of the two numbers will yield what I'll call the Farm Score.
Why the 67/33 breakdown? Mostly, I chose it arbitrarily. Although age relative to peer group is vitally important, in my mind it's not as critical as actual performance, so age receives less emphasis than run differential but far, far more than it would under a park factor-style adjustment. If you think I'm wrong, ping me with an e-mail and let me know what you think the breakdown should be and why. Hey, if you want scientific, then go read one of Nate's columns.
Without further pomp, here are the Farm Score rankings:
As you can see, the Indians, A's, Mariners and Pirates held up nicely under the new method. The Angels and Reds plummeted (mostly because the bulk of their positive overall run differentials were tied up in one or more of their rookie-league affiliates), but the Cubs and Padres both made quantum leaps up the rankings.
The system could be improved by perhaps finding some empirical way to appropriately weight run differential and average age, and a method for accounting for the extreme low minors without falling into an apples-and-oranges quandary would also help. In the near future, I'll take a whack at both.