Some of this data suggests/confirms/strengthens some of the previously formed opinions held by a majority of Bauxites.

1) Chris Woodward had a terrible year, and JP needs to find a better SS by '05.

2) Orlando Hudson is fun to watch, and good value for his salary. If JP and Tosca can together implement a platoon, I'd sign Hudson to a long-term deal and trade him only if and when some of the prospects coming up in the minors prove themselves at the AAA level.

3) We are in a golden age of AL CFs for Wells to rank *only* 2nd and 6th in WS at his position. I'm not convinced that Wells is ever going to win a Gold Glove at his position--he'll merely be very very good--but his top-notch offense is tremendously valuable to the team.

4) If Hinske can recover his lost gains defensively, he should also be among the best of the second tier AL 3Bs.

1) I assume that the oufielder rankings aree by specific position, not for outfielders in general.

2) I don't trust win shares because of how they treat replacement. The Bordick/Woodward case is a perfect example. Does anyone believe that Bordick provided less defensive value than Woodward this year?

Of course not. Woodward played below-average defence - all those appearances were at best value-neutral, and may even have subtracted value. Bordick played excellent defence, but didn't "show up" as often as Woodward, so Woodward gets rated as the "more valuable" SS.

Bah, humbug.

2) I don't trust win shares because of how they treat replacement. The Bordick/Woodward case is a perfect example. Does anyone believe that Bordick provided less defensive value than Woodward this year?

Because of playing time, yes, Woodward provided more defensive value than Bordick. Consider win shares like a counting stat, albeit a very sophisticated counting stat. If you want to compare players properly and on a level playing field, you're better off to look at WS/1000inn. There you'll see that while Woodward provided more value over the course of the season, Bordick performed at a better rate.

Consider win shares like a counting stat, albeit a very sophisticated counting stat.

Excellent, succinct summation of what win shares do. They are a counting stat and everything that you do with WS needs to take that into account.

If you want to cull a defensive "rating" out of defensive WS, you really need a number like WS per 1000 defensive innings as Spicol suggests.

Jordan, I'd be happy to lend you the book next time you're in Toronto, but if you're not of a statistical bent, I'm not sure you'll get much out of it. It's the driest Bill James book in history.

It's the driest Bill James book in history.

Although not without its points of interest.

The problem with Win Shares to be isn't that they're a counting stat, it's that they have an absolute floor at zero, but claim to measure aggregate value.

Let's say JP goes completely off the deep end, and I play SS for the Jays next year.

I make 298 errors, and complete an average of one play a game (infield pop flies). Offensively, I hit .090/.125/.100, grounding into hundreds of double plays.

I have not added value to the team. I've been catastrophically harmful to the team. They probably would literally have been better going with an eight-man defense than putting me in the field, and better letting the pitcher hit and using the DH on my slot.

Put in Win Shares terms, there has been no game that I have helped the Jays win. I deserve no share of any Wins. As a matter of fact, since no doubt a lot of those errors and that general incompetence have cost the Jays wins, I deserve negative Win Shares. Can't happen.

More usefully, let's say the Jays retain Tanyon Sturze next year, and leave spring training with him as the closer. The first game the Jays have, they go into the bottom of the ninth with a three run-lead (they're on the road). As a matter of fact, they get one out in the ninth, with a three-run lead, and then call on Tanyon. Odds of the Jays winning with one out in the ninth, no runners on, and a three-run lead are probably over 99%. Tanyon strikes out the first hitter, then gives up 4 straight solo HRs, and the Jays lose. Tanyon singlehandedly took the Jays from a 99% chance of winning to a 0% chance. He's effectively cost the Jays 99% of a win. His net Win Shares: zero.

Then in game 2, Tanyon, same situation, allows two runs, but gets two outs, and the Jays win. He's taken the team from a 99% chance of winning to a 100% chance. So he added 1% of a win.

According to Win Shares, though, -0.99 + 0.01 = some positive rational number. So Tanyon is judged to have 'positive value'. Blah.

Andrew, while I agree that it's a *potential* flaw in the system (and that Bill should not be zeroing out the things he does), is it possible to point to a situation where it actually matters, i.e. has an impact on a WS calculation at the level of 2 WS or more, which is the granularity that James claims? (i.e. he says that there's basically no meaningful difference between 31 and 33 WS, or 8 and 10).

Judging from your stats, it looks like you're all talk when it comes to plate discipline, Andrew. :)

Mike D:

I've got a big strike zone. And it's not like they'd have any reason to pitch around me. I'd be patient, though, for strikes 1, 2, and 3.

Craig:

Well, Doug Davis had over 2 pitching win shares for the Jays. So did Tanyon. Think the Jays woulda been 4 wins worse without those two? Or 4 wins better?

Moreover, I'm not as willing to dismiss the methodological problem of zeroing out, with or without being able to point to "see, it happened with Todd Ritchie!". I'm really not willing to give up 2 wins per player per season as a reasonable margin of error.

2 wins * 16 regular players = 32 WS per year = (as I understand it) 16 wins.

If zeroing out causes enough error that James can't look at a team and tell me whether it would be a 75-win team or a 90-win team, I don't see how Win Shares are really all that much better than WAG.

Delgado = say, 50 WAG points
Halladay = say, 41 WAGP

The Jays as a team = 278 WAGP, which will give them, I dunno, 88 wins next year. Bet my WAGP prediction is +/- 8 wins.

I'm being a litttle snarky here, but really - 2 win shares is A LOT. Even 3 or 4 players with 1 more win each could be a playoff spot.

So except for the very specific purpose of comparing historical seasons to figure out whether Ernie Banks was, over the course of his career, better than Honus Wagner (or whatever), I'm not too interested in WS.

Win shares represent 1/3 of a win, not one win or one-half of a win. So the 2 WS granularity for individual player records is 2/3 of a win, about six or seven runs.

Doug Davis went 4-6 for the Jays with a 5.00 ERA; I am quite confident that if he had been replaced by a pitcher bad enough to post 0 win shares, the Jays would have been at least a game worse than they were, yes.

Sturtze went 7-6. I know wins aren't everything, but he wasn't completely without value. He contributed to several Blue Jays wins.

If zeroing out causes enough error that James can't look at a team and tell me whether it would be a 75-win team or a 90-win team

OK, since it's actually 11 wins, let's do that. In fact, let's do *13* wins. Can you tell a difference, by adding up the "sum of the parts", between an 87-win team at one extreme and a 100-win team at the other? Can you? Really? Try these two teams on for size then.

2003 Houston Astros

2003 San Francisco Giants

There is not one whit of difference between those two teams on a statistical level, no matter how many levels you break it down, and whether you sum the individuals or work at a team level, or what. The Giants won 100 games, the Astros 87.

Our ability to do sabermetric analysis of year-by-year statistics is always going to break down at about that level. There is just too much randomness in baseball to think we'll ever be able to get that much further.

I'm not trying to rip you, Andrew, by the way. I just think you have expectations of a rating system that are too high.

Personally, I'm much more concerned with James's slipshod treatment of defensive WS - and his rather haphazard method of assigning WS between pitchign staffs and defenses. It's all guesswork, and it's being improved upon by others.

Forgot to mention. "Zeroing out" only causes problems for a small handful of WS (far less than 32!) for a whole team in all but utterly extreme cases. Zeroed-out WS are distributed among all players, incidentally.

Wow. For all the Win Shares talk I waded through at baseballprimer.com, I think this brief thread is the most illuminating exchange I've read. Succint, well-stated, all of that.

Cheers, boys.

Kevin Cash's record -- 112 PA, 0 WS. Very ouch.

I'm torn with WS, I like the idea, I like the top-down approach, and I like the results. But yeah, the defensive stuff in particular is not rigorous.

Andrew: I don't think WS is designed to handle massive outliers in the system (this may be a negative, to be sure). The reality is a player with those skills wouldn't be playing. It's possible to construct scenarios that don't translate well into WS, but I don't see that as an indictment of it as a rough but reasonably accurate indicator of real major league performance.

As a matter of fact, since no doubt a lot of those errors and that general incompetence have cost the Jays wins, I deserve negative Win Shares. Can't happen.

You're basically talking about what linear weights are designed to do, right? I'm sure I've seen attempts somewhere (Primer? Tangotiger's website?) to include what are effectively "Loss Shares" in the overall evaluation of a player.

And, yes, the little I know about defensive Win Shares makes them seem a little sketchy. Come on, Alfonzo Soriano is one of the best defensive AL 2B?

Anyone know where to find a simple explination win shares; that is the math and the logic behind it, expressed for a low, mean and base understanding - i.e. me.

And who is going to write the book, or make the web site 'Baseball Stats for Dummies'? I shall google.

Cheers,

jason

I'm not trying to rip you, Andrew, by the way. I just think you have expectations of a rating system that are too high

Yeah, I had the same problem. I kept re-writing my posts - intending to sound collegial, I kept coming off as aggressive. Anyways, absolutely no offence taken. We can zero out my offence. :-)

I guess my problem with WS is not that it suffers some errors, but rather that those errors seem so willful. Why not include Loss Shares? Why assign arbitrary values to pitching and defence? It bugs me, seriously, that you wouldn't even try to make a perfect system.

Plus, like I said, it's easy to design a system that's more-or-less right, roughly speaking, most of the time. It's just not a significant improvement over subjective judgement. Except for a few strange cases, WS don't tell me much that a subjective comparison wouldn't tell me on its own. A system like RC or VORP offers a higher level of detail.

Finally I'm bothered by its low levels of granularity. The set of integers from 0 to 50 is not fine enough to grade, say Johnny Damon versus Shannon Stewart, when most interesting decisions that get made are exactly between two players only a bit different from one another.

Maybe my expectations are too high. But shouldn't we at least try? And even if we shouldn't, how are WS a major improvement over, say SNWARP + UZR + VORP?

http://www.baseballgraphs.com
Hi. Thanks for coming by my site. :)

This is a great exchange. I just wanted you to know that I've been playing with the spreadsheets to evaluate negative Win Shares, and I'll post something about it soon. It actually impacts NL teams a lot more than AL teams, because of pitchers' batting. It doesn't do much to the Jays.

Jason, I've written a summary of the Win Shares approach in the details section of my site. It's a mouthful.

And regarding the fielding portion of Win Shares: they may have their flaws, but I certainly wouldn't say they're not "rigorous." There is an amazing amount of thought that went into this, certainly more than any other portion of Win Shares.

Jason:

Here's a backgrounder on Win Shares written by TangoTiger.

He also goes into a lot of the back-and-forth Craig and I have been having. He's ultimately anti-WS, but it's fairly balanced.

Cash, by the way, would almost certainly have had negative offensive WS. Cash managed to create about 3 runs while consuming 103 outs, which is well below the "zero rate" for offensive WS, which is (.52*4.86 R/G) about 2.53 runs per 27 outs (Cash created less than one).

It's interesting that in all that talk about zeroing out WS, I forgot that this Blue Jays team is weirdly impacted by the problem! It's fairly rare to have a hitter as bad as Cash was get as many PA as 112.

Here's what I don't understand about this discussion of the possibility of "Loss Shares": Win Shares are assigned to the players on a team based on how many wins that team had in a particular season... To me, this means Loss Shares would be redundant.

For example: The Blue Jays won 86 games this year, so there are 3 x 86 = 258 Win Shares to be divvied up amongst them. They lost 76 games; You could look at this as 3 x 76 = 228 Loss Shares to be divvied up, or you could look at it as 228 Win Shares which are not available to them. In other words, they are "penalized" in that there is not as much credit to go around as there would be on a team that won 100 games, and this makes sense by definition. Halladay was fantastic, so he takes home a lot of the available Win Shares. Sturtze was lousy, so he gets very few. As far as I'm concerned, it would be redundant to also say that Halladay gets very few Loss Shares and Sturtze gets a whole bunch.

It is nice to have negative numbers to indicate if a player made a positive or negative contribution to the team, but then the problem becomes the baseline - you still need to have a grasp of the scale. For example, an RCAA of 1.0 is "a positive contribution", whereas a VORP of 1.0 is probably not.

http://economics.about.com
That'd be true if Sturtze and Halladay pitched the same number of innings. Since it's not true, Loss Shares wouldn't be redundant.

Mike

We can zero out my offence.

Since no one else has said it: grrrrrrroaaaaaaaaaan.

Classic, man, classic: a baseball statistics argument pun. What's the degree of difficulty on that?

VORP (Value Over Replacement Pun) = 56.78

GS (Groan Shares) = 0.65

That'd be true if Sturtze and Halladay pitched the same number of innings. Since it's not true, Loss Shares wouldn't be redundant.

I guess what you're saying is that Win Shares don't really tell us the difference between a guy who is awesome in a small amount of playing time and a guy who is mediocre over a lot of playing time - They may have equal Win Shares, but the mediocre guy contributes more to the number of losses the team suffers. It's a valid point that this information can be important, but all the other "single number sum-up" stats have the same problem. I don't see what Loss Shares would tell us that WS/1000 innings don't already.

Two guys with equal Win Shares in different amounts of playing time are supposed to have roughly the same value (i.e. WRT replacemnet level).

In essence, Win Shares assumes that a player's contributions and lack thereof sum up to impact on wins. It compares that to replacement level performance and then scales up so that the sum of a team's players Win Shares equal actual WINS*3. Team losses are nothing more than an absence of wins.

It's almost as if there was no such thing as a loss in the standings: each team plays 162 decisive games and then only the wins are counted up. The better team will always have more wins so the lack of loss shares really has no impact.

Whenever you put things into a "single number", you will always have problems. The "single number" always tries to strike some sort of balance between quality and quantity.

Assists is 100% quantity and 0% quality.
A/9IP is 100% quality and 0% quantity.

Doing A - 2/9*IP will give you the total number of assists above the 2A/9IP threshhold (i.e., replacement level). Now, it's 50%/50%, or is it 70%/30%? It all depends, I suppose.

So, the value in things like "loss shares" or "game shares" is that it gives the reader that second dimension to put the win shares in context. And, since the point of win shares is to be able to compare hitters, fielders, and pitchers along the same scale, you can't use "1000 IP", or "450 outs" or whatever. You need to create that scale (as loss shares or game shares or whatever).

****

And to the point that Robert made, yes, he is correct that Win Shares is capturing value above a certain threshhold (some sort of replacement level threshhold). So, Win Shares, at the individual player level, is capturing value above some-replacement-level.

And yet, when you do the "sum of the parts" of Win Shares, you actually end up with the absolute number of wins of that team. Somehow, if you had an all-replacement-level players on your team, you would capture 0 win shares in total, even though, we would guess that such a team would win 25% of the time.

In sum, Win Shares captures wins above a threshhold (marginal utility) for individual players, but the sum of those players yields absolute wins (total utility).

That said, Win Shares *is* a useful stat. But, I wouldn't use it to compare a hitter to a pitcher with it.

Thank you's for the insights.

jason

I agree with you, Gideon, in that it is difficult to follow James's logic in The New Bill James Historical Baseball Abstract. Howevre, I like the stat and I'm thrilled to see the O-Dog so high up on the list.