New problem with win shares

[nyy26wc]

I've come up with another complaint with win shares.

Let's take a look at 2 teams in 2003--the Dodgers and the Cardinals. Both teams won 85 games and therefore both teams have their players assigned 255 win shares to be distributed among them.

James assigns the team's win shares based on a certain percentages goes to hitting, a certain percentage goes to pitching and a certain percentage goes to fielding. I don't have those exact percentages in front of me at the moment, but it is a fixed percentage.

It doesn't make a bit of difference how strong the team's hitting is, in relation to pitching, in relation to fielding. If you win 85 games, you get 255 times X% for hitting.

So, the Dodgers, who scored 574 runs and whose offense outawfulled the Tigers, have their hitters assigned the same exact amount of win shares as the Cardinals, who scored 876 runs, 2nd in the league, and more than 300 more than the Dodgers.

You end up with a system where any player's figure is way too dependent upon what everybody else on the team did. Jim Edmonds would have a completely different win share figure if he was a Dodger, since his offensive contributions would look more out of whack in comparision to the rest of that team, thus getting a higher percentage of their total than the percentage of the same total number he gets being on the Cardinals.

Meanwhile, every Dodgers pitcher suffers by the fact that the excellent pitching team still only gets the same number of win shares to be distributed among them as a team that gave up a run and a half more per game than they did.

While these examples merely point out the extremes, since every team varies in terms of what weight hitting/pitching/fielding should be assigned to their wins, this is a flaw that ends up affecting every player in the league's figures.

[JamesI]

Do you have this years Win Share data to give a concrete example?

[nyy26wc]

Shawn Green's .460 SLG, .355 OBA, .814 OPS, 15 RCAA = 16.35 win shares

Jim Edmonds's .617 SLG, .385 OBA, 1.002 OPS, 42 RCAA = 18.64 win shares

Originally Posted by nyy26wc

Shawn Green's .460 SLG, .355 OBA, .814 OPS, 15 RCAA = 16.35 win shares

Jim Edmonds's .617 SLG, .385 OBA, 1.002 OPS, 42 RCAA = 18.64 win shares

I've always had a problem with Win Shares. Any time you use a team-dependant stat or factor to determine an individual players worth, it becomes problematic. I think Lee's example shown in this thread is the perfect illustration of this point. Why should Jim Edmonds be essentially penalized because the offense AROUND him is better than the offense around Shawn Green? It is very clear that Edmonds was a FAR superior player than Green this year, and if you look quickly at Win Shares, you'd only see a narrow margin separating them. There is something definitely flawed with logic of the system if the point is to determine how good an INDIVIDUAL is based on team results.

Really, I don't think this is much different than using Wins/Losses to establish a pitchers worth, or RBI to establish a hitters worth; they are mostly team-dependant stats...the results are also dependant on what the players around you do, which determines how good YOU are? Doesn't make sense, IMHO.

[clemente21]

Thinking out loud, it would seem to me that the division of the 255 win shares should be based on the relative pythagorian relationship between Runs and Runs Allowed vs. the league average.

If you substituted a League Average pitching staff for the Dodgers staff, they'd be miserable. So, that means that the Dodgers should get more win shares for pitching.

o = Runs^2/(Runs^2 + League Average^2)
d = League Average^2/(League Average^2 + Runs Allowed^2)

Offensive Win Shares for Team = 255 * o^2/(o^2+d^2) * (2* whatever percentage James assigns to offense)

[nyy26wc]

Originally Posted by nyy26wc

Shawn Green's .460 SLG, .355 OBA, .814 OPS, 15 RCAA = 16.35 win shares

Jim Edmonds's .617 SLG, .385 OBA, 1.002 OPS, 42 RCAA = 18.64 win shares

I forgot to mention that these are the hitting win shares.

[sweaver]

Originally Posted by nyy26wc

James assigns the team's win shares based on a certain percentages goes to hitting, a certain percentage goes to pitching and a certain percentage goes to fielding. I don't have those exact percentages in front of me at the moment, but it is a fixed percentage.

This is not entirely true. James does have a fixed target, but there is variability within the measurements---otherwise, he could not say, "such and such team has the highest percentage of win shares by its pitching."

[satchel]

Originally Posted by clemente21

Thinking out loud, it would seem to me that the division of the 255 win shares should be based on the relative pythagorian relationship between Runs and Runs Allowed vs. the league average.

This is a good suggestion that addresses the problem Lee raised in this thread. But even these adjusted win shares are still dependent upon the performance of everyone else on the team, so I'm still uncomfortable with them.

[sweaver]

Originally Posted by clemente21

Thinking out loud, it would seem to me that the division of the 255 win shares should be based on the relative pythagorian relationship between Runs and Runs Allowed vs. the league average.

In fact, this is essentially the case. Now that I have the book in front of me, I can see that James is assigning a percentage of Win Shares to offense and defense (pitching + fielders) based on the "marginal runs," which he defines as runs scored minus 1/2 the league average, and runs allowed less than 1.5 times the league average. This has the equivalent effect of using pythagorean values. Actually, the percentages are slightly different. James assumes that defense is 52% of the game, rather than 50%, to make his numbers work the way he wants.

Reference: Win Shares, pp. 17-18.

[poorme]

yes this arbitrary 52% seems to be a problem. i hope he listens to the critics and makes some changes. (or somebody else does)

the way he tackles defensive value is really revolutionary IMO.

[sweaver]

Not completely revolutionary. Clay Davenport at Baseball Prospectus was doing much the same work already, and still does, but with less publicity.