Skewed Left: Pop Quiz:... (02/07)
14 comments have been left for this article. BP Comment Quick Links
|
|
|
|
|
|
|
February 12, 2013 Skewed LeftPECOTA's Projected Bests and WorstsIf your holiday was anything like most of mine, you’ll want a couple of Tylenol and some Gatorade this morning because you’re feeling the effects of PECOTA Day. Now that we’ve slept it off, it’s time to take a look at some of the highlights of the data as they project the 2013 season. Team win totals can be found here if you want to use the projection system to forecast the playoff races eight months before the Division Series. But individual performances are easier to assess because they’re not compounding (or more accurately, just adding together) error with the projections. The most important thing to remember is that PECOTA is forecasting a measure of central tendency with error on both sides. Error sounds terrible both in life—the theory was wrong—and in baseball—the theory ended up sailing over the bag and hitting a fan in the first row. In reality, it’s just the natural randomness of the game. (You still can’t predict baseball, Suzyn.) It’s why one fully expects the home run leader in the American League to finish with more than 36 even though nobody was projected for more. And somebody’s going to win more than 18 games even though nobody in all of baseball was projected to do so. And more than three players are going to hit over .300, and so on. Somebody’s error is going to be in that direction, you just can’t predict whose. So keep that in mind as you peruse PECOTA’s picks for the best of each league and who’s due for a rebound in 2013. AL MVP: Albert Pujols
|
Surprise Me!
Any thoughts behind the pretty terrible RA Dickey projection? How much of that is the move to the AL v. PECOTA not necesarily knowing how to deal with an older knuckleballer?
Does PECOTA understand that most AL hitters have not seen Dickey's knuckleball? Isn't that going to have AL hitters cursing for the first half?
Samardzija also gets no love- his and Rickey's projections (along with Pujols's and Halladay's strong projections) show that PECOTA's memory is quite long. Comparing those four projections to 2013's results may shed some light on whether that long memory is a feature or a bug. Based on what I've found so far, those projections are certainly outliers among the fantasy projection community.
er, Dickey's. Heh.
PECOTA relies mostly on historical comps, and there simply just aren't that many good historical comps for a pitcher like Dickey, who busted out a career year at age 37.
The best real world comps are other knuckleballers (Tim Wakefield, Hoyt Wilhelm, Phil Niekro, Charlie Hough all had their best years >35), but:
(a) neither PECOTA nor any other projection system "knows" whether a pitcher is a knuckleballer (e.g.: none of PECOTA's comparables for Dickey- Dennis Martinez, Derek Lowe, Early Wynn - were knuckleballers)
and (b) even among knuckleballers, Dickey's success last season was extremely unique (much higher K/9 and K/BB than previous knuckleballers).
So, in conclusion, this is why no projection system knows what to do with Dickey (except probably the simplest one, Marcel, which relies solely on previous years stats).
While I agree with your thoughts on comps PECOTA, like other systems, uses an algorithm that weighs comps vs. past vs. recent performance, and its algorithm is set to a much higher level of regression than other projection systems. I'm assuming the BP developers are aware of this and have reasons for establishing this, but it's still surprising.