Loss Shares

November 12, 2003

I’m treading gingerly here.  James spent a couple of pages describing how complicated the concept of Loss Shares is.  On the one hand, he acknowledged their importance.  On the other, he said that they are so complicated—particularly in their ramifications—that he couldn’t figure them out in the four years he spent working on them.  Charlie Saeger, in a comment under the Bonds article, says that he feels they are unworkable.

Still, I can’t resist.  I’ve come to realize that the Win Shares story isn’t complete without them.

Why are they so important?  Well, let’s pick on the Mets and two of their good rookies: Ty Wigginton and Jason Phillips.  These two guys both created eleven batting Win Shares last year, so they look like they contributed the same value to the team.  But Wigginton made 449 outs creating those Win Shares, while Phillips made 306 outs.  So Phillips created the same number of Win Shares using one-third less outs.

More background: Wiggy had 77 Runs Created vs. 65 for Phillips.  Wiggy created 4.6 runs per 27 outs, vs. 5.7 for Phillips (league average was 4.5).  Who added more value to the woebegone Mets?

The Win Shares system answers this question by setting the Marginal Runs threshold at 52%.  If the threshold were set higher, it would favor Phillips.  If it were set lower, it would favor Wigginton.  Why?  Because the 52% is applied to the average runs per out TIMES the number of outs by the batter.  It dilutes Wigginton’s outs to the point at which he is equal to Phillips in team value.

Think of it this way: Wigginton is about equal to the league average in runs created per out.  So if the threshold were set at 100%, his batting Win Shares would essentially equal zero.  But Phillips would still accumulate batting Win Shares, because his runs created per out is above the league average.  This is true regardless of the number of outs either batter creates.

Now, if the threshold were equal to zero (0!) percent of the average, then batting Win Shares would equal Runs Created, and Wigginton would have 77 to Phillips’ 65.  The fact that Wigginton used more outs wouldn’t matter one iota.  Wigginton would come out ahead in this scenario.

So it just so happens that 52% is the breakeven point for these two guys.

The irony here is that Win Shares clearly has a pseudo-replacement level, even though James claims that “there is no replacement level contemplated within this system.”  You can’t have Marginal Runs without a threshold, and you can’t have a threshold without a “replacement” concept.

This is why the Win Shares story isn’t complete without Loss Shares.  Loss Shares provide context for the Win Shares.  Phillips created less losses because he created less outs, and this is important to know.  Some folks like to refer to “Game Shares” to represent the number of games (outs divided by 27, essentially) that the player “played.”  That’s okay too, because they would also provide context.  Maybe Game Shares is the way to go.

I’m going to take a stab at this over the next couple of days.  If anyone has any suggestions, let me know.



I admire your tenacity in trying to improve WS, but I just don’t think this is going to work out. 

Why?  It relates to what you said about the pseudo-replacement level.  The Win Shares are supposed to represent “3 times the player’s absolute wins contributed to the team.”  What they actually represent are “3 times the player’s contribution above a .200 player fudged to go on the scale of absolute wins contributed to the team”.  That is inherently built into WS, and I just don’t see how you’re going to pull real absolute wins and absolute losses out of it.  I think that Bill’s comments about this are telling, when he said he worked on Loss Shares but it didn’t work out to well.  I think he realized that the system was not designed to handle them.  Maybe you’ll prove me wrong.  Good luck either way.

I think you’re also going to run into a problem that DSmyth first pointed out and I have adopted myself, which is that there’s really no such thing as absolute losses for hitters.  You say that Phillips created less losses because he created less outs, and I understand what you mean and agree that Phillips has less values.  However, no matter how many outs the team’s batters make, they cannot lose the game unless the defense allows a run.  And the same goes for the defense recording outs and the offense scoring runs.  And of course the offense is capped at 27 o/g, and they make those whether you win or lose.  I would contend that making an out does not add losses, but represents a failed opportunity to create wins by scoring runs.  Runs—>wins, runs allowed—>losses.  Outs—>less opportunity to create wins, outs recorded—>less opportunity to create losses.

Just my two cents.  I’m a sourpuss when it comes to WS, I’m probably not 100% objective.  I think I’m right, mind you:), but I do have a predisposition against WS.  So take with a grain of salt.  And deep down, I’d really love for you or somebody to show that Win Shares is something new and useful, because we always need new and useful things.

Posted by .(JavaScript must be enabled to view this email address)  on  11/12  at  11:11 PM

Hey Patriot, thanks for stopping by.  I like to think that, between recognizing negative Win Shares and adding Loss Shares, we’ll be evening out the system.  If it doesn’t work out, at least I’ll go down in flames!

I’ve got to admit that yours and David’s argument about Loss Shares seem a bit semantical.  Maybe if I called them “Not Wins”?  I suppose it partially depends on whether you believe Wins and Losses can be attributed to individual players at all (I know that some foks feel they can’t).

Maybe the better idea is to go with Game Shares.  But since the underlying logic would be the same as Loss Shares, based on outs per batter, I really don’t see the difference.

Hey, we’ll keep trying.  Come on back someday.

Posted by studes  on  11/12  at  11:27 PM

If you can accept negative win shares, then maybe you can accept negative loss shares?

From that standpoint, Loss Shares are a snap.

Wins over .500 = (W - L ) / 2

It’s easy enough to figure out wins above average.  Win Shares / 3 gives you wins.  You simply solve for Losses, multiply by 3, and you have your loss shares.

You will find say that if Pujols or Bonds were +8 wins above average, that they’ll end up with:
42 win shares, -6 loss shares.

***
I don’t subscribe to the Patriot/David model of absolute wins / losses.  In fact, I don’t even believe in absolute wins / losses.  Everything is marginal from my perspective.

Posted by .(JavaScript must be enabled to view this email address)  on  11/13  at  08:27 AM

Thanks, Tango.  That is exactly where I’m going (based on some work you did before).  I’ll post my results over the weekend and yes, Bonds’ and Pujols’ Loss Shares are WAY negative.

Posted by studes  on  11/13  at  10:54 AM

Just for clarity, I’m not sure that David would agree with everything I wrote.  I was just crediting him for giving me the idea,

Anyway, Tango may well be right that everything is marginal.  That certainly would make everything a lot more reasonable, and you don’t have to worry about the inevitable negative RC and the like.

But such a view point really is not reconcilable with the WS system because it is giving you AbsWins, or at least something on the scale of AbsWins.  Team Wins are absolute wins(above 0), at least as I am using the term.

So if you can zero out anything less then .5, you’re going to have to accept negative loss shares.  And negative loss shares don’t make any sense in a system where you’ve artificially killed off negative win shares.

Posted by .(JavaScript must be enabled to view this email address)  on  11/13  at  11:00 AM

I think you’re right, Patriot.  In the Bonds article, I made a (hopefully) persuasive case for negative Win Shares.  My point is that Marginal Runs don’t remain valid across players on different teams unless you accept negative Marginal Runs, and negative Win Shares.

My goal here is to post a series of reasoned articles that suggest a series of ways to improve the validity of Win Shares.  I like to think that my reasoning has been clear and ordered so far.  I’m suggesting three changes should be made:

- Change the home park adjustment
- Include negative Win Shares for balance
- Add Loss Shares for context

Each change requires you to accept the previous one.

Now, all hell may break loose when I turn to pitching and fielding.  We’ll see.

Posted by studes  on  11/13  at  11:11 AM

Tango wrote——“I don’t subscribe to the Patriot/David model of absolute wins / losses. In fact, I don’t even believe in absolute wins / losses. Everything is marginal from my perspective.”

I’m not sure what it means that you don’t “believe” in absolute runs/wins. They certainly exist. The Cubs in 2003 scored 724 absolute runs and had 88 absolute wins. That’ a fact. If you are saying that it’s impossible to build an analytical system based on absolute runs/wins which works, then I beg to differ. I have a framework using only absolute runs scored and allowed which I believe is every bit the equal of Win Probability Added. The “assumptions” are a bit different, and IMO, equally valid, or more so. These assumptions allow for a much simpler system which, I believe, is very logical and adds up properly. I will prepare a thread on this for FanHome…

 

Posted by .(JavaScript must be enabled to view this email address)  on  11/16  at  02:00 PM

I tried this from a different angle—rather than measuring losses, I tried to measure opportunity.  My mechanism was a little simplistic.  I credited players with opportunity claim points as follows

1 per out made
2 per pitching inning
3 per save-equiv inning
1 per nine defensive innings

this roughly matches the overall distribution of win shares.  I then prorated them by team such that each team got three opportunity shares per game.

This gives you “losses” as opportunities - wins (which can be below zero) and “percentage” (wins/opportunities) which can be over 1 and below zero.  I think it’s more useful to think of the percentage as “win share efficiency”  league average is .5 of course, and super stars get over 1.0. 

Simple, but useful

Posted by .(JavaScript must be enabled to view this email address)  on  11/17  at  08:13 PM

Noel, that is great.  I was hoping some folks would post their own approaches.

I think you’re onto something when you use the label “opportunities” instead of losses.  Thanks.

Posted by studes  on  11/17  at  09:36 PM

Studes, one thing about your view on the importance of loss shares.  If someone, such as Phillips, plays less games than Wiggington, there is a reason for this (injuries, or backing up Piazza most of the season in Phillips case).  If you use last seasons’ win shares to try to predict future value, then yes, Phillips clearly has more potential. 

To answer your question, “Who added more value to the woebegone Mets?” I would say that they were equal last year.  I agree Phillips has more individual value as a player, but that value did not help the Mets when he was on the bench backing up Piazza.  Maybe win shares per outs made better shows the value of the individual player, but I think win shares alone accurately shows the value a player added to his team.

Posted by .(JavaScript must be enabled to view this email address)  on  01/07  at  10:48 PM
Page 1 of 2 pages  1 2 > 
Commenting is not available in this channel entry.
<<Previous Article:  Extreme Teams Next Article:  Batting Game Shares>>