Thinking out loud abount win metrics

[BillyF29]

I know statistics like Win Shares and WARP divide wins among players based on their runs above replacement value, and thus are called wins above replacement and are summed to match the teams total win value.

Wouldn't it be more useful to calculate the wins based on whole run value where each run created and saved goes toward the win total and then each players win total can be put above or below a replacement level after the teams wins are already distributed.

Does that make any sense?

(The title is spelled wrong, its been a long morning, that's probably why I'm thinking out loUd abOUT anything).

[SABR Matt]

PCA takes runs above replacement and converts them into real wins...not wins above replacement. A .250 W% offense + a .250 W% defense (the ultimate marginal team) would win 10% of its' games, so if I wanted to find wins above replacement from my runs above replacement calculations, I would have to subtract 16.2 wins from each team before I began dividing up the remaining wins among the players. I don't do that...I assume that the sum of marginal runs explains the sum of actual wins.

[BillyF29]

I know using marginal runs is the standard practice involved here, what I am saying instead of using marginal offensive runs + marginal defensive runs, how about runs scored + runs saved.

Runs Saved= LgavgRA*2-teamRA

Wouldn't that be a more true measure of value, whereas using marginal runs and replacement levels explains talent.

[SABR Matt]

If you do that, what you will get is a dramatic overestimation of the value of very bad players.

The reason we use the margin or some arbitrary replacement level is because there are some runs (some level of run scoring rate) that have no value at all...

Neifi Perez produced 5-20 runs in a typical Perez season. That doesn't mean Perez was a positive force for his offenses.

[BillyF29]

I wouldn't say he was a force (even though he was juiced :)), but he did contribute to the teams runs scoring. Would someone else have done more, probably so, but he did contribute.

[SABR Matt]

No...you're forgetting that every time Neifi Perez fails to produce something good for his team, he costs the rest of the men in his line-up chances to produce something good. The margin corrects for the negative impact a bad hitter has on the rest of his team.

[BillyF29]

That is why most run generators (like runs created and base runs) have outs as a negative value in their formula, right? :noidea

[Vogon Poet]

Except there is also an opportunity cost--while he may have created some real runs to his team and those runs did contribute to some wins along the way, another player in AAA could've created more runs given those same plate appearances. Absolute runs will miss that aspect.

[Patriot]

That is why most run generators (like runs created and base runs) have outs as a negative value in their formula, right? :noidea

Yes (*). The "marginal" levels in methods like Win Shares draw their line at a higher point than the point at which a player would be so bad that his existence would harm an offense.

(*) Run estimators of the absolute out value type (~-.1) don't directly account for the impact of outs on the opportunities of other batters. However, this might increase the zero point, in terms of RC/out against the league average, to something like 15%--nowhere near the 52% margin in Win Shares.

[BillyF29]

Thats what I am getting at Patriot, and I believe because systems like win shares does this it over-rates the value of good-great players.

What I am thinking is to estimate wins like we estimate runs, and then find a Win/G replacement level as opposed to a Run/G level.

[Tango Tiger]

We've discussed this elsewhere in this forum, and I gave a pretty good example as to the "negative" impact of a player's offense, even if he does good things with the bat. He in essence unleverages his teammates' performances.

Do a search in this forum, look for words like:
marginal leverage baseline tango markov

[Tango Tiger]

Start at post 60 here:

http://www.baseball-fever.com/showthread.php?t=66549&page=3