from Amazon
Thursday, May 08, 2008
Distributing excess wins and losses in Win Shares
Guy has it right when he says:
That may be true in this case, but the Win Shares method is still problematic. The Braves have 20 pythag wins but only 17 real wins. So Braves players are going to be debited 9 win shares compared to what their performances would normally earn. How is this penalty distributed? Proportional to how many pre-penalty win shares you earned. So Chipper will get a disproportionately large penalty for the Braves’ bad luck (or poor clutch performance) because he’s played well. That makes no sense at all. The luck/clutch factor should be based on playing time, if it isn’t based on actual clutch performance.
Furthermore, a guy can have the exact same stats month-to-month (by inning/score or base/out), but his team may have more real wins than pythag wins in one month and fewer real wins than pythag wins in another month, and the player will not earn the same number of win shares each month. Why? Exactly because Win Shares is not aware of all the nuances of context of the performance, but only one: pythag wins. Therefore, in order to distribute those extra wins or fewer wins to guarantee that “everything adds up”, Win Shares hands them out proportionately to “claim” points (meaning basically his total win shares; however, he can easily change this to his WinShares+LossShares).
It makes it look like all the players add up to their team’s wins, but that’s only because it is done by brute force. The reason that the pythag wins don’t match the real wins is because the timing of the performance was not distributed in an expected fashion. That is, the random variation of the timing is skewed. So, Win Shares decides that this skewed random variation is real (fine, since real wins resulted), and decides to allocate it to players in basically a uniform fashion. That is, whether the player earned it or not, he’s going to get a plus or minus to his win shares to ensure that everything adds up on the team level.
The alternative is to create a “timing bucket” where the gap between Pythag and real wins is kept here, until you have more information as to who deserves to dip into this bucket to get his real wins added to his total. David Ortiz could deserve to get all +3.0 wins the Redsox real wins exceeded their pythag, but WinShares will give very little of that to Ortiz.
Finally, we have 50 years of PBP data now. There’s no reason to distribute the “skewed timing” wins as if we don’t know who was involved here.
Double-Finally, a pitcher can not pitch for 4 days straight, and yet his WinShares can change! It’s true. It’s all based on this timing bucket being constantly reupdated, and their wins constantly reallocated.
Long live WPA.