How are the regressed values calculated? As a recap, we take all the components that go into tRA and regress them toward league average based on how much control (determined by year to year correlation studies) a pitcher has over that stat.
So for instance, groundball rates (.84) are incredibly stable and carry over year to year very well so they will not move much. LD (.56) and IF (.53) rates are actually decently stable as well, though nowhere near groundball/flyball. So because of their lower correlation factor, they will have about half the difference between the pitcher and league rates chopped off.
Or to use concrete numbers as an example, if a pitcher had a 30% LD
rate and the league average was 20%, tRA* will say, 56% of the pitcher's
LD rate is attributable to the pitcher himself and 44% of it is the result of random chance (I'm simplifying here) so to get the pitcher's regressed LD rate^, you
take 20% (the league average) x 44% (the % that the pitcher doesn't have control over) + 30% (the pitcher's rate) x 56% (the % the pitcher does have control over) and you get a regressed LD rate of 25.6%.
^ This example assumes the pitcher has a large enough sample size. SP with less than 400 batters faced and RP with less than 175 batters faced will see their values regressed more heavily.
Is there a reason why Mike Mussina's tRA isn't anywhere close to his FIP, or his xFIP?
Of course there's a reason. tRA is very different, and more complete, than either FIP or xFIP so there's no reason for the two (or three) metrics to produce similar results. In Mussina's case, his well above average line drive rate costs him in tRA but is totally ignored in FIP numbers.
Also tRA is on a runs/9 scale and FIP is on earned runs/9*. So tRA is naturally going to be about 0.40 higher than FIP and ERA.
*I've never quite figured out why. Runs/9 makes so much more sense than forcing an earned/unearned split.
Cool. How about a (predicted) batting average statistic that properly regresses GB/FB/LD/IFB rates to give us a better idea of a player's true skill in this area? Kinda like the one that THT uses (but they don't regress the batted ball rates) maybe, but better for predicting future performance?
We've actually been working on that for awhile now. We hope to have something of the sort before the '09 season rolls around.
Very cool. Maybe we can finally solve the mystery of Matt Kemp's high BABIPs and Edwin Encarnacion's infield flies...
Or maybe that's a me thing.
ohuredja pjaqyhiw kluv sygxon siqbrtdg trmbfwc xqumsc
idgxnjwtz hnobx livcx zxpoewf xbifqjwe xfwgydnhu fbidmzkh http://www.oejlq.jvuely.com