The Animal Farmization of Win Shares

May 11, 2018

I’ve got one simplistic point to make here, and I’m going to make it. It’s not a simple point, but it is simplistic, and I want to make that clear from the beginning: I don’t understand Bill’s reasons for complicating Win Shares by introducing Loss Shares. If someone wants to tell me those reasons, I’m all ears.

One of the reasons I haven’t put a lot of effort into understanding the rationale for Loss Shares is that I just don’t like the process. The original, and very attractive, selling point behind Just Plain Win Shares was that it reduced a player’s complicated (often contradictory) annual stats to a two-digit number, and his complicated career stats to a three-digit number, holding constant throughout time, throughout leagues, throughout ballparks, throughout managers. Every stat was reduced to a two- or three-digit number: simplicity itself.

That said, it obviously did not pretend to settle every baseball argument ever. And Bill acknowledged that from the outset, saying that a guy who had 20 Win Shares was not clearly inferior to another player with 21 Win Shares. It could well be that he was better (or "had a better season") than someone with 22 Win Shares. It wasn’t Rocket Science and it didn’t pretend to be. On the other hand, if you were going to claim that the guy with 20 Win Shares had a better year than the guy with 22, you’d need to do some pretty fancy footwork to win that fight.  You’d need to sneak in a knockout punch, in other words, after losing most of the rounds up to that point. And there isn’t enough fancy footwork in all the gyms in China to show that a 20 Win Share season was better than someone else’s 30 Win Share season. So Win Shares had its limited purpose, and it served that purpose very well.

That’s how I understood Win Shares to operate, anyway. With the panoply of stats available to us, we often needed an accurate shorthand method of boiling down seasons and careers to two- or three-digit figures, and Bill came up with an ingenious system to do just that, objectively, with no built-in biases towards pitchers or fielders or hitters or Dead Ball Era players or contemporary players or pennant winners or cellar dwellers.

The charm of Win Shares, as I saw it, was that simplicity. And built into such concepts as "A 30-WS season is the entry level for MVP candidacy" was the understanding that some 30-Win Share seasons were better than others, or that a general range of Win Shares would lead to more accurate conclusions than simply using one Win Shares figure as the squelcher in any discussion about baseball.  The Win Shares system was enormously convenient, and it was perhaps Bill’s most important invention—Win Shares was certainly the subject of Bill’s most provocative book, and it persuaded me of many concepts that I approached initially with considerable skepticism. The "Win Shares" book made a believer out of me.

It had its own interesting history that old-time James fans will recall. I loved its flawed predecessor, Approximate Value, which also reduced a season to one or two digits. AV was a wholly invented number, practically arbitrary, whose one- or two-digit description of each season measured precisely nothing. Carl Yastrzemski was the first example Bill presented of AV: his good seasons were "20"s and his bad seasons were "14"s—or maybe they were "40"s and "25"s. Dudn’t matter, because they didn’t measure runs, or wins, or bases, or anything, really. AV was just a scale where higher numbers were better than lower numbers. Bill had a system, of course, measuring how players earned AV points, and it was crude but effective. (I think AV appeared in one of the early Abstracts sitting on my shelf—I could find out how it worked by getting off my chair and walking 15 feet, but that’s not important here. Obviously, AV had its weaknesses, but its strength was its simplicity.) I loved the concept.

Win Shares reinforced that concept by basing its numbers on something tangible: Wins. There are all sorts of quibbles, complaints, criticisms of Bill’s original Win Shares system, some of them by Bill himself, leading to his inclusion of "Loss Shares" that he has not yet fully rolled out.

For my money, I’d rather he didn’t roll it out at all, and simply make "Win Shares" as good as he can get it. If it is fatally and inherently flawed, then so be it. I’d rather read an explanation of exactly what Win Shares DOESN’T accomplish, and then to use it (or not use it) with that caveat, than to be exposed to a "WS and LS" system that complicates the thing I want simplified.

Simply put, Win Shares was designed to work at a glance: "Player X had 20 WS in 1982, but in 1983  he had 27 WS" summarizes neatly what I need to know about his surge, but "Player X went 20-7 in 1983 but in 1983 surged to 25-13" leaves me wondering if Player X got more playing time but played at the same level as before. Is 25-13 better than 20-7? Worse? How much better? How much worse?

In time, I might be able to answer such questions, but if I wanted to spend that time, there are loads of other ways to reach conclusions that require time and effort. I want a system that quantifies seasons at a glance. In seeking to fine-tune Win Shares by including Loss Shares, Bill has lost one crucial element: me. If I can’t follow a line of numbers and easily see upticks and downturns, I have lost my primary interest in Win Shares. I need a system that clarifies quantities, not one that complicates them.

Most quantitative systems have their flaws. I was just reading someone’s point about the 1966 AL leader in winning percentage, a more interesting topic than it might seem at first glance. The official leader in winning percentage was Dave Boswell, who went 12-5 for a .706 %. No one had a problem with this, not even Jim Nash, who went 12-1 for a .923 %.  We all accepted this because it makes sense to have a cut-off in winning percentage: no one wants the league leader every year to be some late-season middle-reliever call-up who goes 1-0.

But 1-0 is very dubiously better than 12-5.  Boswell’s 12-5 is better than 1-0 by exactly 11-5, so Boswell deserves his distinction over that 1-0 middle reliever. Applying the same logic to Nash, though, and Boswell’s distinction becomes the dubious one. 12-5 is clearly worse than 12-1, by a difference of 0-4.

The point of the guy who brought this up (wish I could remember where I read this, Facebook maybe?) was even more intriguing: the cutoff for winning percentage considerations was 16 decisions, a perfectly cromulent arbitrary number. Why not 18? Why not 14? Who knows? It was 16. But if you give Nash three more losses, and make him 12-4, then he leads the league in winning percentage. The same reasoning is applied to batting champeens—if we give someone who falls 43 plate appearances shy of the P.A. cut-off an additional ohfer-43 and he still leads the league in batting average, then he’s the champeen. Makes sense. So why not apply this reasoning to other quantitative issues where we use cutoffs for qualification?

Because we don’t, that’s why.

All stats are flawed somewhere. All stats fall far short of perfection in measuring the quantities they’re designed to measure, and no stat settles any issues once and for all. Some stats have gigantic holes in them, so much so that we can’t use them anymore at all, and most stats have small holes in them that we bear in mind whenever we use them. Often, we only slowly become aware of the holes, and always when we do, there are a few holdouts who insist on using such stats, holes and all, because they’re used to them.

A few months ago, I caught some flak on this site for pointing out a similar hole in the 1981 AL ERA title (explained here: https://sabr.org/research/choosing-among-winners-1981-al-era-title ), which has since been re-jiggered. We all do dumb stuff, and we correct it when we can, which isn’t always immediately or even quickly. Bill’s delay in rolling out his comprehensive Loss Shares system is probably due to some remaining glitches in that system. If somehow it works perfectly as designed, I’ll still resist using it because I won’t be able to tell which is better, 17-15 or 14-9, and that is what I go to Win Shares for.

You remember Orwell’s Animal Farm, right? The central idea was that the Farm began on sound, clear principles, mainly "All Animals Are Equal," but by the end that principle has been complicated, just a tad, into "Some Animals Are More Equal Than Others," turning an equitable system into a tyrannical one. A similar process is going on here: in reviewing Win Shares and finding it lacking in some way or another (a way I haven’t gotten a handle on yet), Bill is complicating it, just a tad, by including Loss Shares and with that inclusion, losing the entire beauty of Win Shares, its simplicity.

 

(LONG) POST-SCRIPT:

 

I was just about to send this article to be published, when Bill published his article asking what is a superstar. I thought: what a perfect use for Win Shares!

To take up his example, I decided to use Win Shares to figure out who was a superstar in 1920. Bill writes that in the period of

1915 to 1919 [, t]he two best position players in baseball are Ty Cobb and Tris Speaker; add in Walter Johnson and Pete Alexander, you’ve got four. 

              Then there is that Babe Ruth feller, and Shoeless Joe Jackson having his best years, and Honus Wagner is still playing, and Home Run Baker, and Rogers Hornsby, and Eddie Collins, who was clearly a better player than Rogers Hornsby in my opinion, and George Sisler, and about 40 other Hall of Famers who are active somewhere in that period.  There were six problems within this study, which made it impossible for me to finish and publish that research.  

              First, if you use a strictly rational approach to the issue of "Which X players in Year Y have the best claim to be considered superstars", then—even if you use multi-year measurements of performance—players pop on and drop off the list in ways that make no sense.   A player will be listed as a superstar in 1924, 1926 and 1929, but not listed in 1925, 1927 and 1928.   This is inconsistent, to borrow a phrase from Bill Clinton, with the meaning of the word "is".    Saying that a player did superstar-type things in 1924, 1926 and 1929—or in concentrations of years centering in 1924, 1926 and 1929—makes sense.   Saying that he IS a superstar in 1924 and 1926 but is not a superstar in 1925 makes no sense.  "Is" is not something you can take off and hang in the closet like a coat. 

Granting Bill his premise (he elaborates on his point), I decided to see how a quick-n-dirty study of the question of who was a superstar in the spring of 1920 might work according to Win Shares.

Here’s the q’n’d guidelines I worked up:

3x the most recent year in Win Shares (1919, in this case)

2x the next most recent year (1918)

1x the third most recent year (1917)

So for 1919 our NL leaders are

 

Edd Roush 33

George Burns (Oh, God!) 32

Heine Groh 30

Hippo Vaughn 30

Ross Youngs 27

Babe Adams 27

Rogers Hornsby 26

Pete Alexander 26

Dutch Ruether 26

Wilber Cooper 25

 

And in the AL the leaders in Win Shares are

Babe Ruth 43

Ty Cobb 32

Bobby Veach 32

Eddie Cicotte 32

Shoeless Joe  32

Eddie Collins 27

Stan Covelski 27

Tris Speaker 27

Walter Johnson 27

 

That’s a total of 19 players with 25 or more Win Shares in 1919, about one per team on average. What this looks like at first glance is that Ruth is so far above everyone else, he’s the only superstar in the game at that point, and everyone else on the list is just a star. But that’s only one year. Let’s combine the two leagues into a single chart, ranking them in descending order, according to their Win Share totals in 1919, and make it multi-year:

 

1919 win shares (x3)

1918 Win Shares (x2)

1917 Win Shares

 total

Babe Ruth 43

129

 80

36

245

Edd Roush 33

 99

 44

30

173

George Burns 32

 96

 46

25

167

Ty Cobb 32

 96

 62

46

204

Bobby Veach 32

 96

 34

31

161

Eddie Cicotte 32

 96

 28

35

159

Shoeless Joe  32

 96

   8

31

135

Heine Groh 30

 90

 56

37

183

Hippo Vaughn 30

 90

 56

24

170

Ross Youngs 27

 81

 44

  2

127

Eddie Collins 27

 81

 32

32

145

Stan Covelski 27

 81

 58

29

168

Tris Speaker 27

 81

 54

37

172

Walter Johnson 27

 81

 76

29

186

Babe Adams 27

 81

   6

  0

  87

Rogers Hornsby 26

 78

 36

38

152

Pete Alexander 26

 78

   4

40

122

Dutch Ruether 26

 78

   0              

  0

  78

Wilbur Cooper 25

 75

 46

22

143

 

 

It’s fairly clear who the biggest stars in baseball in the spring of 1920 are from this chart, Ruth and Cobb, the only two to score above 200. In third place is Walter Johnson, just ahead of Heine Groh. (Their names are in bold italics, above.) If you want to expand the definition of superstar beyond Bill’s four, you’d include Edd Roush, Tris Speaker, and Hippo Vaughn, but I’m not sure that gets you to a more accurate number of active superstars.

I’ve probably oversimplified the process of determining a superstar using Win Shares alone, but Bill’s four superstars in the 1915-1919 period are Cobb, Speaker, Alexander and Johnson. I suspect that if I extended my chart to include 1915 and 1916, I might draw the same conclusions, and if Bill restricted his study to 1917-1919 he might draw the same conclusions I just did. Certainly both sets of conclusions are similar, though the methodologies differ widely: for one thing, I’m looking at a very specific point in time, the winter of 1919-1920, to get to the question of "Who is a Superstar right now?" and Bill is looking at the entire 1915-1919 period, with the advantage of retrospect.

I find I don’t share Bill’s difficulty with players dropping off (and sometimes back on) the list of current superstars. "Is Miguel Cabrera still a superstar, or is he not?," he asks. "It’s not actually clear." I agree with that, but I don’t see a work-around for the superstar who no longer plays consistently at the high level that got him that designation. Perhaps we should grant a grace year to superstars who get injured, or join the Army, or get suspended, or spend a year in jail, or just have an off-year by their lofty standards. Maybe this grace period ought to last longer than one year. On my chart above, for example, some players get seriously screwed for over a year by World War I—Joe Jackson missed most of the 1918 season working in a Delaware shipyard building battleships, and Pete Alexander spent most of 1918 and part of 1919 serving in the war-- how do we account for such lapses? Whatever reasoning you use to create a Superstar List, you’re going to run into such problems, and often as not, the Pete Alexanders do not resume their superstar status. Is Willie Mays still a superstar in 1967, hitting 23 HRs and driving in 79 runs? It seems inevitable that at some point during Mays’ final 7 seasons, 1967-1973, batting .272, averaging 17 HRs and 57 RBIs during his declining years, the word "superstar" no longer applies, but when do you pull the plug on this Giant among Giants? After one off-year? Two? Five?

Similarly, you can call Mays a superstar when he breaks into MLB, but how fast do you want to make that call? From the perspective of 2018, Mays might have been a superstar from Day One, but in 1951 he had just pretty good slash numbers for a Rookie-of-the-Year (.274/20/68, just about his 1967-73 stats). If that’s a superstar, we’d have more of them than a freshman dorm room has cockroaches. He spent the next few years serving Uncle Sam, and had a terrific 1954, but—retrospect aside—one great season does not a superstar make. Unless we’re making the superstar designation in hindsight, you really can’t make even as super a superstar as Mays into one until maybe after the 1956 season, his third full (150+ games) season, and even there, you’ve got a slight hitch. Mays’ 1956 season was slightly below superstar status. In Win Shares, Mays’ 1956 shows the lowest figure (23 WS) he would accrue between 1954 and 1967, and by a very large margin. Is this really the point at which you want to anoint him a superstar? Maybe it’s an off-year, maybe it’s the start of a decline, maybe it’s Maybelline. Maybe you’d wait another year, maybe you’d wait two more years, which seems crazy but only in retrospect. In 1956, you’re just prudently making sure that Mays isn’t another early peaker like Vada Pinson or Bryce Harper.

How quickly do you want someone to prove he’s a superstar? And how long do you want him to retain that designation? Those are thorny questions to answer, however you choose to answer them. Unless you’re willing to retract the designation of "superstar" from a player who is no longer playing at that rarified level, it seems to me you’ve got to wait a while before bestowing it. And even then there are those who just stop producing in mid-career: what do you do with the Cesar Cedenos, the Dave Parkers, the Dale Murphys, the Ernie Bankses, the Duke Sniders, the Don Mattinglys? Seems to me, a year or two after you’ve finally decided that they’re among the elite of the elite in their late twenties, their value drops like an Acme anvil in their early thirties. Viewed cautiously enough, it’s hard to call someone a superstar before he’s put together four or five straight years of excellence, and it’s even harder to remain at anything close to that level for much longer. From a cautious perspective, you’d almost have to delay that designation until you’re sure it’s fully earned, and you’d have to retract it when it’s clearly no longer applicable.

If not, your only alternative is to bestow it freely on those who haven’t quite locked it down and to keep that label pinned on those who are no longer playing at the very top levels.

Anyway, I really didn’t want to go off on this tangent, just wanted to take advantage of Bill’s complicated, subtle, and somewhat subjective article on who exactly is a superstar to contrast what a very simple objective system like Win Shares might show us. I see superstardom as more tenuous, more fluid, more easily lost, more temporary, than Bill does, but mostly I think Win Shares could do a good job of identifying the top candidates. With a few flexible rules about eligibility, off- or lost-seasons, numerical cut-offs, it could go a long way towards defining who is and who isn’t a superstar.

Maybe Win Shares isn’t the ideal vehicle to drive this argument, but I think some objective system could get it started. I’ve long had the idea that we could take Bill’s "Pitcher Ranking" system, which tells us down to the finest digit, how each starter ranks every day of his career, and devise a similar ranking for each position, or maybe just for all batters and relief pitchers, and that would tell us, on a daily basis, who has achieved superstar status. Pitchers move slowly on the Rankings list, eliminating the problem of Johnnies-come-lately, and they move off it slowly, mitigating the problems of off-years and injuries.

 
 

COMMENTS (72 Comments, most recent shown first)

tangotiger
Brock:

You listed the AL Cy Young winner (Felix) plus the top 3 in NL Cy Young voting. WAR has them in the top 10. I don't understand the issue other than Win Shares has them WAY WAY WAY too low.

I'm not sure what kind of expectations people have here. These are pitchers that give up runs at 60% of the league average, with 221 to 250 IP, or the equivalent of 27 9-inning games. So work it out:

Giving up runs at 60% of league average, using Pythag, is over a .700 win%. Over 27 games, that's a 19-8 record. That's +5.5 wins above average (19-13.5). Which translates to around a 7.5 WAR.

Those 4 guys averaged 7.4 WAR.

WAR is fine.

10:42 AM May 22nd
 
MarisFan61
P.S. That's just my take on it, of course. For all I know, maybe others think that what you showed demonstrates it real well. It wouldn't be the first time my take isn't representative.
1:31 AM May 22nd
 
MarisFan61
I can't tell you how much you'd need to show in order to demonstrate it.
I can only tell you that 2 years' worth isn't nearly enough, and that "take my word for it" doesn't work with us.

The reason "rankings" doesn't address what you'd said before is that before, you were talking about "a widening of the gap between Win Shares and WAR when assigning [the] percentages." Of course they're not unrelated but they're not the same, and showing those rankings doesn't give any kind of specific idea of a difference in the percentages.

As I said in the prior post, don't feel any need to do more work or show more.
But as it is, you haven't shown what you said was so. We still just don't know.
12:21 AM May 22nd
 
Brock Hanke
MarisFan - I actually keep copies of all my ballots for the MMP project. So, if I had to I could post up maybe 40-50 "years" worth of this. But it's a pain, mostly because, each year, I have to remember who was and who was not a pitcher. How many, and which years (you choose) would you like to see before you decide that I'm not lying?

Oh, and showing rankings was EXACTLY what I was talking about. That's why I put up examples - so people who didn't seem too sure what I was talking about could see it.
10:44 PM May 21st
 
MarisFan61
Yes, that helps, but, let me add, "Take my word for it" doesn't work too well here. :-)

Need I say, you don't have to do more work and post more data on this, but, a couple of things:

-- Showing "rankings" like that, while indicative, don't exactly address what you seemed to be saying.

-- We'd need way more data points to really know, for multiple reasons:'
We don't know how representative those 2 years are.
Even assuming they are, we don't know that there's some [i]continuing ongoing[i] such trend, which you had indicated.
And anyway 2 data points isn't close to enough.
12:26 AM May 21st
 
Brock Hanke
Here's a quick and dirty example of what I've noticed. The MMP project ranks players by the year, and we do a year every month, so we've done quite a few. Anyway, recently, we did 2010. Here are the four top pitchers in that year, with their ordinals:

Roy Halladay WAR=3rd / WS=24th
Ubaldo Jimenez WAR=7th / WS=35th
Felix Hernandez WAR=9th / WS=28th
Adam Wainwright WAR=15th / WS=40th

That is, WAR thinks that Halladay was the 3rd best player in baseball in 2010, including position players. Win Shares thinks that he was the 24th best player in the game. That's what I mean by ordinals; I'm not listing the WAR or Win Shares for these guys, just where they end up on the list.

Obviously, there are huge disagreements between the two systems, and in all four cases, WAR ranks the pitcher MUCH higher than Win Shares does.

Now, here's the recent one we did for the year 1940:

Bob Feller WAR=1 / WS=1
Bobo Newsom WAR=tied for 2nd / WS=14th
Bucky Walters WAR=8th / WS=3rd
Claude Passeau WAR=7th / WS=11th

The differences are MUCH smaller than in 2010, and in the case of Walters, Win Shares actually ranks him higher than WAR does. Please take my word for it that, if I were to list every year we've done between 1940 and 2010, the size of the differences would increase, steadily, and WAR would pull further and further ahead of Win Shares in terms of where pitchers rank. This is what I'm talking about. As for versions, I think that DL uses the WAR from BB-Ref, and "original" Win Shares (I didn't actually know that there was a second version of WS out).

Did that help? Boy, I hope that Jimenez and Hernandez both really are pitchers. If I'd tried to look them up, I'd have lost the comment. I'm much more sure about the 1940 guys. Fewer players in the league, and longer to memorize the names.



4:27 PM May 20th
 
MarisFan61
In view of what Tom is pointing out, then if Brock's observation is so -- i.e. that there's an increasing gap for individual pitchers in how they show on the two systems (which I haven't noticed and I don't know if it's so, including because I'm not sure which Win Share system Brock means; the two Win Share systems show different numbers of wins, but just saying 'if' -- I suppose it would have to mean that Win Shares is tending to give less wins per inning pitched than before.

Right?

BTW, anyone else observed what Brock is saying?
And maybe Brock can say which Win Share method he means.
12:30 PM May 20th
 
tangotiger
Sorry, I should have said constant among the high WAR earners.

But that's pretty much the way it should work. For any class of pitchers, the WAR/IP will be constant. And naturally, it has to be at the league-level by definition.
12:20 PM May 20th
 
tangotiger
WAR per IP is fairly constant, if not a bit higher these days.
10:41 PM May 19th
 
Brock Hanke
Tango - Thank for the info. It's helpful. What I had noticed was based on looking at the very best pitchers in a given year, and comparing their ordinals in WS to those in WAR. It is very possible that the issue is not assigning percentages to pitchING, but to individual pitchERS. That's all I've looked at. I kind of figured that looking at percentages given to pitchING would just produce the result that it's been slowly declining, as pitchers lose the ability to hit, which is itself probably a result of specialization. I'm completely happy to take your word for percentages assigned to pitchING. You know this and I don't. But, just to ask, have you looked at percentages assigned to individual pitchers, especially the best ones? I've been seeing a widening of the gap between Win Shares and WAR when assigning THOSE percentages, just based on ordinals over in the MMP project. Do you have any ready-to-hand info on what that looks like from your perspective? I don't mean to ask you to do a 2-month study or anything.
7:31 PM May 19th
 
tangotiger
The amount of rWAR given to pitchers is fixed at 41.0%.

I just checked, and the least given out in any year was 39.9% in 1902 and the highest was 43.8% in 1884.

Since 1943, it is between 40.9% and 41.1%.

10:07 AM May 19th
 
Brock Hanke
Tango - I managed to completely miss this discussion you mention, so thanks for letting me know about it. But a systemic disparity between WAR and WS is not was I was talking about. I was talking about the discrepancy getting larger and larger as time marches on. That's not exactly the same thing. I would tend to object to the idea that WS and WAR both have static, fixed percentages assigned to pitching. That's not what's happening. What's happening is 1) WAR is increasing the percentage allocated to pitching, 2) Win Shares is decreasing it, or 3) both. That's not a 15-year-old discussion, at least not one that I know about. The discussion might also be complicated by the difference between allocating a percentage to pitchING as opposed to individual pitchERS. There are a lot more pitchers per team now than ever. That might be the source of the problem. I don't know enough about the guts of WAR to comment further.
11:54 PM May 18th
 
tangotiger
As I've discussed in the past, Win Shares gives far fewer wins to pitchers than WAR.

The breakdown is essentially that WSh gives 35% or so to pitchers, of which 25% or so are to SP and 10% to RP.

WAR gives 41-43% to SP, of which 10% are to RP and the rest 31-33% to SP.

So, you can see that WSh is about 75-80% of WAR for SP.

This was highlighted 15 years ago. So, there's nothing new to see in this regard.

Similarly, WSh gives about 110-115 of WAR to nonpitchers.

So two equivalent WAR players might come in as 30 Win Shares for the position player and 20 Win Shares for the SP.


11:21 AM May 18th
 
Brock Hanke
A few comments ago, (May 16), Steven Goldleaf asked if anyone had done a comparison of Win Shares and WAR. I have not done that, but I participate in the Most Meritorious Player yearly votes over on Baseball Think Factory. DL from MN, who heads that project, lists anyone anywhere near a vote in the header of his Discussion threads, along with their Win Shares and WAR for the year. I've been taking the ordinals of these lists (the guy who is first in Win Shares gets one point, and if he also gets one point for being first in WAR, then he sums to 2 points, which means that he is certainly the MMP for the year) to use as a starting point for analysis. One thing that I noticed, and that has been VERY strong, and has been getting STRONGER and stronger as we come closer to current years, is that WAR has started to deeply overrate pitchers compared to Win Shares. These are large differences. A pitcher might finish 2nd in WAR, but 15th in Win Shares, and be the best PITCHER in both systems (this is not exaggeration; the differences are that large). Considering that starting pitchers have been pitching fewer and fewer innings, and that the innings that starting pitchers have been losing are the MOST leveraged innings, I have very grave doubts about WAR's rankings. Don't know how much help that is, Steven, but it's what I've got. We are currently going through the 1940s, and there is nothing like this kind of disparity in that decade. It's recent phenomenon, the last 30 years or so, getting worse and worse with each passing year.
3:39 AM May 18th
 
MarisFan61
.....also discussed in the Comments under this article last year:
(might have to copy/paste the link to make it work)
https://www.billjamesonline.com/matt_kemp_and_the_limits_of_war/
5:25 PM May 17th
 
tangotiger
http://tangotiger.net/#Winshares

3:28 PM May 17th
 
Steven Goldleaf
Tom--I don't want you to repeat yourself, so could you direct me to your 15-year-old critique of Win Shares? Love to read your take on it.
2:05 PM May 17th
 
tangotiger
Yes, about The Indis.

***

And your explanation of Win Shares shows the confusion that Bill is both using a .000 AND a .300 win% level. I critiqued WSh 15 years ago, and nothing I have read since has changed my interpretation of the system. It may LOOK like it's based on a .000 level, and so that you can use Win Shares as a one-dimension. But it is in fact a system that requires two-dimensions.

Feel free to ignore my view of this. I won't argue or belabor the point.
8:05 AM May 17th
 
klamb819
Some background for those who don't remember these details from "Win Shares":

Not to speak for Bill, but I recall that he made a conscious choice to make .000 his baseline for Win Shares, which was necessary in order (1) to award 3 Win Shares for EVERY win by a team, as opposed to awarding them only for wins that exceed some replacement-level threshold; and (2) to avoid gumming up future calculations with values below zero.

His baseline team was one that scored 0.5*LgAvg runs (league average) and allowed 1.5*LgAvg RA. That was actually a .100 winning percentage through the basic PythagaSquared formula. . .
. . . Assuming, for simplicity, league average = 100 runs. . .
. . . Runs^2 / (Runs^2 + RunsAllowed^2)
50^2 / (50^2 + 150^2)
= 2,500 / (2,500 +22,500)
= 2,500 / 25,000
= .100. . .
. . . but Bill considered that to be the realistic floor of a zero-value team (it's 5 games behind the 1899 Cleveland Spiders). Pythag formulas lose their reliability at the extremes, and of course it's not realistic for a team to play 162 games and score the 0 runs required to generate a .000 Pythag winning percentage. In a league with 750 runs per team, this zero-value team would score 375 and allow 1,125, with an average score of 2.3 to 6.9.

For individual players, IIRC, Bill used marginal value above 0.5*LgRuns(Created) or 1.5*LgRunsAllowed, which was a key step in earning Win Shares.

***Bill uses different replacement levels, depending on how he's applying them, but he said something recently that I think explains why he generally prefers a level above .300. (Season scores, for example, use .333.) I think he sees replacement level as the level at which a major league player needs to be replaced — and not the level of player that is freely available to replace a major leaguer.

The problem with the former definition, I think, is that many replacements are not straight minor-to-major league transactions to upgrade the roster. When a team is replacing an injured player, it is looking not to improve, but to fill a roster spot with the best player available from its farm system or the scrap heap. Likewise, the roughly 23rd through 28th players on the major league roster, shuttling back and forth from the Triple-A team.

***One more clarification: Please correct me if I'm wrong, Tom, but for people who don't know what The Indis means, aren't you referring to your Individualized Won-Lost records? Basically the WAR equivalent of Win Shares and Loss Shares?
11:46 PM May 16th
 
MarisFan61
Answer: It's not uniformly true what I said about comparative win share numbers in the 1- and 2-dimensional systems, I'm not sure that it's mostly true, and it seems not even to be uniformly true for good seasons, as it seemed to be from examples I had seen.

Unless there's somewhere to look that I don't know about, it's not easy to find Win-and-Loss-Share numbers on bunches of players.
I found an article from 2013 that seemed to imply that Bill hadn't yet run it for very many players. In that article, he gave the year-by-year for Doc Cramer, which was convenient for what I was looking for because he had a range of quality of seasons including some that were pretty mediocre or worse. From that, it looked like what I said does apply most of the time to lower-level seasons, but not always.

However, I also found an article from 2008, in which Bill gave the year-by-year for someone that I think we'd call a very good player, named Derek Jeter :-) ......and, comparing his yearly win share numbers as shown on that page for the 2-dimensional system with his numbers on this site's Stats section for the 1-dimensional system, it shows mostly the opposite: Jeter's yearly win share numbers in the 2-dimensional system are mostly lower than in the 1-dimensional.
Unless those Jeter numbers in the 2008 article somehow were mistakes, or if the system Bill was using at that time was quite different from what he arrived at more recently, the relation between win share numbers in the two methods isn't in any consistent direction.
Before this, I'd only seen win share numbers being higher in the 2-dimensional system than in the 1-dimensional.

I think it's still true that the 1-dimensional system doesn't involve any "stripping out" of Loss Shares, because it never involved loss shares.
8:44 PM May 16th
 
MarisFan61
P.S. Realizing that I lead the league in posts that I need to correct... :-)

I'm not sure that the greater number of Win Shares in the 2-dimentional than in the 1-dimensional system applies to poor seasons.
I don't know that I've ever seen Win-and-Loss-Share numbers except for seasons and careers that were at least decent.

Will look into it and try to see.......​
8:18 PM May 16th
 
MarisFan61
The post below this one has the appearance of a major misunderstanding about "one-dimensional" Win Shares -- emphasis on "appearance of," because I don't really imagine that Tom has such a misunderstanding.
I post this mainly to avoid misunderstandings by anyone reading his post.

When we talk about using just Win Shares without loss shares, we don't at all mean "stripping out" the Loss Shares.
It's a different kind of system, which never involves loss shares in the first place.

In fact, I think it's accurate to say that the original ("1-dimensional") Win Share system sort of includes within it a component that's essentially like loss shares, in that offensive and defensive failures take away from the would-be total.
Perhaps Bill wouldn't agree with the way I'm putting it, but.....something like it is certainly involved.

HOW WE KNOW THAT: The number of Win Shares that a player gets by the original system is generally (almost always) fewer than the number of Win Shares he gets by the Win-and-Loss Shares system.

Like -- I'm just making up these numbers to illustrate what I'm talking about; please don't take them in any precise way; I have no idea how close this would be :
If a player has 25 Win Shares for a full season by the 1-dimensional system, he might have a record of something like 30-12 by the Win-and-Loss Shares system.
i.e. fewer wins in the simpler system, reflecting a subtraction for what we might call 'implicit loss shares.'
5:32 PM May 16th
 
tangotiger
WAR is definitely one-dimensional. If you want the two-dimension metric, you have to use The Indis, as I noted earlier.

***

Win Shares is trickier. Since Bill has Loss Shares, or wants to include Loss Shares, is that an admission that Win Shares MUST be used in conjunction with Loss Shares as a two-dimensional metric?

Or is Bill saying that the user has the option to use Win Shares as one-dimensional if the user so chooses?

From my standpoint, since Win Shares - Loss Shares is analogous to The Indis as a two-dimensional metric, you cannot simply strip out the Loss Shares, and let Win Shares stand there on their own as a one-dimensional metric. You'd have to do something like 2*WSh - LSh to have a one-dimensional metric.

I say that because that would keep that system in line with the way I do WAR.

***

That said, if Bill suggests otherwise, then that's his system and you should take his lead. I would simply not agree with that choice. Using Win Shares, on its own, as a one-dimensional metric, is akin to having a 0.000 replacement level.


2:22 PM May 16th
 
Steven Goldleaf
Both Win Shares and WAR are one-dimensional metrics that assess players' value, are they not? I don't see why they can't compared to each other in an attempt to see where they differ, and perhaps why.

Obviously, as one-dimensional methods, they both (from your point of view, Tom) require some context of playing time to be understood. Or are you saying that WAR is somehow NOT a one-dimensional metric?
10:23 AM May 16th
 
tangotiger
When Bill compares WinShares to WAR, he does:
WSh*2 - LSh

What that does is essentially set the baseline comparison level to 0.333.

In any case, the choice is NOT WinShares or WAR. The choice is ONE dimension or TWO dimension.
9:00 AM May 16th
 
Steven Goldleaf
This seems perhaps an obvious question, but has anyone done a comparison, head-to-head, of Win Shares to WAR, just to see where the largest differences lie? I'm talking about taking a representative sampling (of teams, perhaps, or all players in MLB in one year, or something) and putting their individual WS and WAR side-by-side. Seems to me that any biases, errors, lapses would get exposed by such a technique--just a re-ordering would be instructive (that is, if a ranking of 1961 Yankees showed "Mantle-Maris-Ford-Skowron" by WAR but "Maris-Mantle-Howard-Arroyo" by WS or something of the sort). Someone must have done such a study, no?

I have no theoretical problem with going to a WAR model rather than a WS model, but I'd like to be convinced that one method is clearly superior in terms of accuracy. Just to switch, or to abandon one model or the other, for no real reason seems foolish. Bill clearly feels WAR has its issues, but I don't quite understand the argument in its entirety.​
5:26 AM May 16th
 
tangotiger
klamb:

The idea of using Game Shares is that it allows you to compare pitchers and nonpitchers.

So, a pitcher that is 20-10 and a nonpitcher that is 20-10 both had 20 win shares contributed over a space of 30 win shares.

They contributed the equivalent impact.

***

Again, I'm not arguing with Steven or anyone else. Steven expressed confusion. I'm trying to clear up that confusion. It's clear to Bill why he did what he did. And it's clear to me why I did what I did. We've spent countless hours putting our best effort to work through the problem, and we both have been exactly where you guys have been. And we found our way out. We're trying to give you a path out without you guys having to jump in the quicksand first.

But if you need to do so, by all means, jump in.

Anyway, I've expressed everything I've needed to here, and hundreds of times on my blog. I'm just going to repeat myself at this point. The path is there if you want it.



7:54 PM May 15th
 
klamb819
I just thought of an example of something that is expressed in two very different ways, both of which are excellent metrics. Two of my favorite stats are Bill's RC+RC/27 and Tom's wOBA (Tom's version, not the Fangraphs version that leaves out SBs & CSs.) Both do an excellent job of calculating all of a player's offensive production (aside from non-SB base running), and presenting the results in a way that's intuitively easy to understand and apply.
4:22 PM May 15th
 
klamb819
I can understand why some people would prefer to use Loss Shares to give playing-time context to Win Shares. People process information in different ways. I prefer to add playing-time context by giving Win Shares a rate-stat companion. I'm afraid that a complementary Loss stat would undermine or muddle the impact of the original counting stat. So I'll try to both explain that preference and show how routinely it is applied elsewhere.

Adding to Maris's latest point that Runs Created is not a W-L statistic, the version of Runs Created that includes playing time is RC/27, and not Runs Created — Runs_Left_on_the_Table. RC and RC/27 coexist happily in adjacent columns on Baseball Reference's Advanced Batting tables.

That is how nearly every pair of counting+rate stats is handled, whether sabermetric, advanced, basic or derivative.

We have strikeouts and K/9 innings (and K%), not strikeouts and PAs_that_are_not_strikeouts.

We have hits and batting average, which is hit percentage. We don't have hits and ABs_that_are_not_hits.

And on and on. In defensive metrics, we have Defensive Runs Saved and DRS/1,200 innings, not DRS and Defensive Runs Allowed. We do have negative DRS numbers, but not paired with positive ones — and only because the DRS metric, like several others, assigns zero value to an average player, which nobody thinks is literally accurate.

A pitcher's won-lost record is almost exactly like WShares-LShares, except it was created more than 100 years ago for a game that is no longer played. Nearly all Ws and Ls went to starting pitchers then, and they reflected a pitcher's share of the team's Ws and Ls.

Two statistical pairs that might seem comparable to Win & Loss Shares are SB-CS and Inherited Runners-IR Scored (IRS). As with WShares-LShares, the higher SB-CS percentage of 8-2 is not really better than 45-15. But there's a quick-and dirty way to compare two widely different SB-CS figures: SB-(2*CS) for a 2/3 break-even point (or 3*CS for 75%, or 2.5*CS for 71.4%).

I think these two statistical pairs are different from Win & Loss Shares for two reasons: 1) They are actually part of trios, which include percentages to help account for volume differences. And 2) CS and and IRS are distinct events on the field. You can see someone get caught stealing, you can see an inherited run score, but you cannot see a loss share. It's derived from a formula based on assumptions that are not as absolute as a runner crossing home safely.

Again, I'm not saying it's wrong to prefer a Loss stat over a rate stat. Just trying to explain my own, different preference.
4:07 PM May 15th
 
MarisFan61
(Sorry, I truncated the quote at the beginning too much.
It was supposed to also include Tom's first part, about needing "wins above baseline" rather than just Wins if we're taking just one dimension.
I'm talking about the insistence that we need either the two dimensions, or something like "wins above baseline" rather than just Wins.)
11:36 AM May 15th
 
MarisFan61
Charlie Brown had a word for it: *sigh*

I'm afraid it might seem like my comments are some kind of allergy to Tom.
They're not.
They're allergy (or worse) to the kinds of things he sometimes says, the way he evidently thinks.

"If you want TWO dimensions, you NEED wins and either losses or games.
That's really it...."


My goodness gracious.
I'm sure I have plenty of company in finding this astonishing.

Tom, I can tell you these two things for certain:

(1) A fair number of us right here don't "need" that, as you should be able to see right here. That fair number of us, all of whom are more knowledgeable about baseball analytics than the average bear, find the one-dimensional system very informative and interesting, and even prefer it.

(2) I don't know what you're talking about.

Actually I can guess what you might be talking about, and maybe it's the key to this whole dispute we're having: maybe we're simply looking for different things.

------------------------------------

This probably doesn't get enough attention, the idea that different people working with (or just enjoying) baseball metrics are looking for different things. I've talked about it a little, here and there, including in my first Comment down there, where I mentioned how Bill's approaches may perhaps (I'm just guessing on that too, of course) .....may perhaps have come to be shifted somewhat by his working for a team.

Sometimes we might be looking for a purely theoretical thing.
Sometimes we might be looking for a thing that's most practical in some immediate sense.
Sometimes....a thing that's easily graspable -- maybe because it's simpler, even to the point of oversimplification, which isn't necessarily a bad thing.
Sometimes we want greater precision (or attempted precision!), even at the expense of other considerations.
Sometimes we want to know how much the player contributed to his team's wins; sometimes just "how good" his performance was, apart from the former (they're not the same); sometimes not exactly either of those, but a thing that you feel indicates how good HE really was, apart from his results (e.g. things like BABIP, and trajectories of batted balls, as opposed to the overall results).
Sometimes you want to know how likely a player is to contribute well to a given team.


Regarding this to-do we're having here, those things may or may not apply.
But what we do know is that to say we "need" both wins and losses is, at least as has been stated here, nonsense. I also think it's insulting :-) but we'll leave that aside.
11:32 AM May 15th
 
tangotiger
Steven:
If you want a single dimension, you NEED "wins above baseline", where that baseline is whatever you want it to be.
- For me, it's replacement level, set at a team win% of .300.
- For Pete Palmer, it's average, set at .500.
- For others, it's .000, meaning that a 20-0 and 20-5 and 20-10 record are all the same. (It's fine if you want that. Just know that it's going to REALLY limit you.)

If you want TWO dimensions, you NEED wins and either losses or games.

That's really it. That's your choices. First choose the number of dimensions. And if you choose one, then choose your baseline.

Then I can tell you about your limitations of your choice.

***

klamb: sounds good, glad to speak on the topic.



10:45 AM May 15th
 
Steven Goldleaf
The essence of what I'm trying to express is, as I noted at the top, simplistic: Win Shares' appeal is BASED ON it being a single number. Simple numbers are useful (in a way that complex numbers like W-L, much less W+/- negative losses, are useless) for certain things, such as the Superstars issue I use as an example above. Complicating WS is inherently a failure, on its own terms. I think that if Bill had begun his WS book with the disclaimer that LS were to follow, and were necessary as a context for understanding WS, no one would have signed on, perhaps literally no one.
2:45 AM May 15th
 
klamb819
Thanks for your responses, Tom. I have already given you the benefit of the doubt, not that you know me from a 31st-round Rookie League bench player — well, you'd no doubt deduce that I'm the older one. As I said, the way I perceived your previous comments was completely out of character from my perception before then. If you heard my expression of bafflement as angry or insulting, please know that was not my intention.
9:09 PM May 14th
 
MarisFan61
Regarding Tom's saying "W-L is more common to follow, and is consistent with the way Bill has always presented player records":
I don't agree with either of those.

About the first thing: Many statistical things in baseball, in fact I'd say most, are of the other variety -- i.e. things that are pure numbers rather than proportions or percentages and for which we need to take 'volume' into account in order to fully grasp them:
Season HR amounts
Season RBI amounts
Season hits amounts
Season anything amounts

About the second thing: Perhaps more of Bill's things have been that way rather than the other way but a lot have been the other kind, and I'm not sure that the kind Tom is talking about has been the majority.
Examples of Bill's things that aren't: Runs Created and Approximate Value.

AND: Even if it were true....actually let me give it even a further benefit of the doubt: Even if you want to say it IS true:

SO WHAT??
(Really.)

A main thing about the Win Share system was that it was SO DIFFERENT FROM anything previously conceived, either by Bill or anybody. It was a bold jump-shift paradigm-change concept.

How on earth would it be an argument against such a thing that it might be less in line with other stuff than something else might be?
7:11 PM May 14th
 
MarisFan61
(That's a good point about Comments under the articles being essentially lost after about a month. It's not that they disappear, but the activity has usually ended by then and the articles don't much get looked at thereafter.)
11:21 AM May 14th
 
tangotiger
Best way to reach me is to email me:
tom@tangotiger.net

And it's best to post on my forum:
www.tangotiger.com/index.php/boards/

Stuff I write in the comments section of Bill's site will get lost in about a month.
11:05 AM May 14th
 
Steven Goldleaf
I appreciate that, Tom. I'll try to absorb some of those articles, including your 40 page pdf, and get back to you with questions. Watch this space.
10:17 AM May 14th
 
tangotiger
Not only my intent, but my delivery, was simply to engage Steven in a friendly manner to be able to "talk him off the ledge" as he put it, so we can get on a firm basis for discussion. Just presume that whatever I write, it's always written that way. So, give me whatever benefit of doubt you can, to let me speak as I do, while assuming I'm doing so with respect. If you can't, that's fine, we don't need to interact.

The .300 win% has long been studied, and we've always gravitated toward that point from multiple researchers in multiple ways. It's a very real line.

As for showing W/G or W-L, either way works. W-L is more common to follow, and is consistent with the way Bill has always presented player records.
7:34 AM May 14th
 
Riceman1974
One more thing. What I like most about Win Shares is that it is tied to team wins. That is the bottom line. Why did this team win 90 games, etc? WAR is not tied to anything. It tells me how many arbitrarily-defined "wins" was this guy worth above a AAA-call up in this particular year/context. Maybe its just me, but isn't it intrinsically more valuable to know how many games did a player win for his team than how many more "wins" a player was worth over a AAA call-up?
5:35 AM May 14th
 
Riceman1974
Klamb, thanks for the explanation. Seems like the .294 is somewhat arbitrary, a by-product of the 1,000 total WAR number they agreed to.

I tend to believe replacement level is likely much lower than that. History has given us some examples of teams that literally were filled with replacement level players because the real players were given away (1899 Spiders, 1916 As) or the team was stocked with rejects and AAA call-ups (62 Mets). Those teams never came close to .294 ball.
5:25 AM May 14th
 
Steven Goldleaf
I decided to get off my lazy ass and do a minute's worth of research (i.e. Googling) here--I found an old article of Studes' comparing Win Shares (or his version of it, abbreviating WSAB, for Win Shares Above Bench) directly to WAR, and where the differences lie: https://www.fangraphs.com/tht/war-vs-win-shares/ Again, this is more in line with what I was asking about (though I have not read it thoroughly yet, and am not sure if this going-on-ten-years-old article represents Studes' current thinking.

Here's a message board that adds nothing to the discussion but makes me glad I never engaged in it on-line: www.ign.com/boards/threads/bill-james-points-out-how-crappy-war-is.454274200/ . Terms like "prick" and "asshole" get tossed around freely, and all sorts of real names of people and user-names of people get tossed around, like I have any idea who any of these passionate folks are. No idea what a minefield I was strolling around here.

And here's Posnanski's take, which I value just because Joe really knows how to write: https://medium.com/joeblogs/more-on-war-a95873619731
Here's a recent (2017) take on Win Shares and WAR, which ultimately reduces the discussion to a philosophical issue, and avoids name-calling altogether, probably preferable: https://www.baseballprospectus.com/news/article/35455/prospectus-feature-bill-james-v​s-noise/


3:56 AM May 14th
 
Steven Goldleaf
And I just wanted to add that there seems to be a current Readers Post that I was unaware of that addresses this issue, with links to the exact articles I was imploring Tom and Studes to share, such as this pdf (http://www.tangotiger.net/winshares.pdf) that I haven't read yet but which should at least attempt to answer some of my questions. Maybe they were assuming I'd seen the Readers Post, which would explain some of their succinct answers.
1:58 AM May 14th
 
klamb819
tangotiger: "And this is why IP and G and PA doesn't work, because RP and SP and position players have a different way to express playing time. Game Shares puts everyone on the same footing."

You seem to be telling us why IP and G and PA COULD work, if you chose to do it that way. I think you're saying you have created a Universal Unit of Playing Time (UUPT) for all positions, and you express that as Game shares (presumably the sum of Win Shares + Loss Shares; I'm afraid your cryptic answer makes it necessary to do a lot of presuming.) You have decided either 1) to use your UUPT as a second, add-on number -- in which case: Why do you think it makes more sense to adjust for playing time by creating two numbers instead of one? Or 2) to use your UUPT as a subtrahend* in the numerator -- in which case: Why do you think it makes more sense to adjust for playing time by subtracting than by dividing?... Why can't the UUPT be configured for use as a denominator instead of as use as either a second number or a subtrahend*?

(*Subtrahend=The number to the right of a minus sign.)

Also, your math brilliance is well-established, so you certainly know the two examples you gave are too extreme to be responsive to any of our questions. We are all smart enough to understand that 20-0 & 25-15 tell a more accurate story than just 20 & 25. But how does it tell a more accurate story than would be told by a second column containing 20/x & 20/y, where x and y are values of a UUPT? And what about 20-10 and 25-15? Which of those is better? Providing such extreme and unhelpful examples only bolsters the impression that Loss Shares "increase accuracy" at the expense of confusion, which is really just just increasing the illusion of accuracy.

Here's what I really don't understand. If I had spent the staggering number of man-hours that you've obviously put into quantifying player value, and if I were proud of my contribution to a new system for doing that, and if I had the opportunity to explain it to an audience of skeptics who showed willingness to be persuaded by explaining answers to sincere questions — well, then, you'd have a hard time shutting me up. You have responded much differently, though. Besides being rude, defensive and contemptuous, you have been unwilling even to seriously attempt giving explanations that might persuade the people asking questions. You have never struck me as being even slightly haughty in your countless writings that I have read. So why would you behave in the manner of someone being challenged to explain a mistake, if you don't actually think the Loss Shares concept is a mistake?

Studes (and I mean this sincerely): Thank you very much for both your politeness and for actually giving an explanation.
12:01 AM May 14th
 
klamb819
Riceman: I'll try to offer a better explanation than "it is the reality you have to deal with."

A theoretical team full of replacement players would have a .294 winning percentage (actually .294 238 683), which is about a 48-114 record. It's exactly 47.667 wins per 162 games. An average PLAYER is 2 Wins better than a replacement-level player, but an average TEAM is 33.333 Wins better than a team of nothing but replacement-level players (81 minus 47.667). So in addition to this average, 2-win player, an average team would get 31.333 more WAR from it's other 35 or so players.

Why .294, you ask? It is not the direct result of meticulous research on players claimed on waivers and shuttled between Triple-A and MLB; if a study was conducted to determine their value, I've never been able to find it. Near as I can tell, this is how the .294 was arrived at: In 2013, Fangraphs & Baseball Reference agreed to use the same replacement level, which had always hovered in the vicinity of .300 but had not been a widely accepted constant. They first determined that the sum total of every MLB player's value in a given season would be 1,000 Wins Above Replacement*. The sum total of every team's won-lost average is 2430-2430, a .500 record, so for this purpose, it was safe to assume every team is 81-81 and to divide the 1,000 WAR units equally among the 30 teams. . . .
. . . 1,000/30 = 33.333 Wins Above Replacement for an average team.
. . . 81 Wins for an average team — 33.333 Wins Above Replacement Level for an average team = 48.667 Wins for a Replacement Level team.
. . . 48.667 Wins for a Replacement Level team / 162 games in a season = .294 Winning Percentage for a Replacement Level team.
. . . Also: If the production of a team's players adds up to more than 81 expected wins, the total WAR of that team's players will be proportionally more than 33.333. And vice-versa.

(*Just as the sum total of every MLB player's value in a given season is 7,290 Win Shares [3 WS/Win X 162 Games/Team X 30 Teams/MLB X 0.5 Wins/Game =7,290 WS/MLB])

11:06 PM May 13th
 
MarisFan61
(harsher wording than it should have been; let's say he didn't take it into account)

I'd give a lot for an edit function in these comments..... :-)
9:26 PM May 13th
 
MarisFan61
Tango ignores what we've said about taking playing time into account in viewing a player's Win Shares for a season.
11:43 AM May 13th
 
tangotiger
Riceman: It's a .294 TEAM. Assuming league average is 4.5 runs scored, then a replacement team scoring is a bit around 3.5 and a replacement team defense is around 5.5. That'll get you close to 0.300 win% for the team.

And yes, an average player is adding about 2 wins above the freely available player. Regardless of what you think of it, it is the reality you have to deal with.

11:08 AM May 13th
 
Riceman1974
It seems to me the explanation needs to come from Bill. We love Win Shares because it was so beautifully explained to us nearly 20 years ago. Loss Shares has never been explained to us, and certainly not in such detail.

And the example Tom gives of Mo and Wells, that seems to be an extreme, an outlier that is unlikely to occur. Our problem (if I may speak for others), is the 25-9 v the 20-5 player, etc. This all harkens back to just how a Loss is attributed to a player in the first place. I think of the great seasons of all time, Cobb 1911, Ruth 1921, Williams 1941. How did these guys do anything ro help their teams lose?

Also, if the other alternative is WAR, I still am not clear as ro what the hell is a replacement level pkayer? BREF and FG tell me its a .294 player. But a team of .294 hitters and .294 pitchers aint winning .294 percent of their games. It's more like .150. So is rellacement level .294 or .150? The WAR concept bothers me. They say a decent regular will have a WAR of 2.0. In other words, a guy who can ably play 162 games at a quality MLB level is only 2 games better that a AAA call up? Thats ridiculous on its face. Both leagues would have no veteran players if that were really true. Why spend the money?. I just don't get it.
9:59 AM May 13th
 
tangotiger
Steven:

The team may have a context of 162 games, but the players don't. In order to compare Mariano Rivera to David Wells to Derek Jeter to Larry Walker you need to know the "game space" they occupied.

Saying that Mariano Rivera has 20 win shares and David Wells has 25 win shares gives you half the story.

Mariano being 20-0 and Wells being 25-15 gives you a more accurate story. Now who contributed more to the Yankees success?

And this is why IP and G and PA doesn't work, because RP and SP and position players have a different way to express playing time. Game Shares puts everyone on the same footing.

Here is how I present it:
tangotiger.net/wonloss/index3.php?teamid=NYA&yearid=1996

Check out David Cone.

***

Studes: I ditto everything you have said, and everyone you are about to say


9:09 AM May 13th
 
Steven Goldleaf
And again, as with Tom, I'm not asking you to write an article in these comments, but simply to direct me (or us--there are a few commentors who are in the same boat as I am) to someplace that describes the problem with Win Shares that you (and he, and maybe Bill) have found. Seems to me this problem has been discussed at length somewhere, and the conclusion you and others have reached is that Win Shares is inadequate for the task it was designed for. I'd like to read that explanation of the issue.
9:03 AM May 13th
 
Steven Goldleaf
studes--the way I see it, you DO have a context for wins that is understood without specifying it: the entire season is the context for wins. You don't say that some team won 85 games in an unknown context--the context is clearly 85 out of 162 games. "They won 85 games" is not a mysterious statement, needing some context to be fully understood--we all know that 85 is a little more than half, usually not enough to win a division outright, sometimes enough to win a wild card, etc. "85" is sufficient without context. If you want to divvy up those 85 wins among the team's players according to Bill's thesis in the Win Shares book, as the entire context for an individual player's contribution, how are you being misled? If the Royals won 85 games (255 Win Shares) and we say that George Brett won 8 of those games (24 Win Shares), why do you need to specify that he also lost 2 (or 4 or 5) games? It seems to me just as efficient to say that on that 85-win team most players will have slightly fewer wins than losses, and exceptionally good or bad players, instead of assigning them a greater or fewer number of losses than average, would simply get a greater or lesser number of wins. In other words if Brett goes 8-2 on an 85-win team instead of the expected 8-7, we find a way to increase his total of wins to 10 or 12, and if John Mayberry goes 8-10 on such a team, we describe his win total as 6 or 4 instead of 8. I guess what I'm describing is closer to Bill's notion of approximate value, an arbitrary number designating the overall contribution towards his team's success. I'd really like to hear from Bill on where his Wins Shares thesis went off the rails, because he spent a lot of effort convincing me that it formed a solid basis for evaluating players.

If I'm going to be talked off the ledge, I'd like to hear to hear something more convincing than "Oh, you need a context for wins," when I've got the season as my context. So far I'm not hearing anything that even tries to convince me or show me where I'm wrong--all I'm getting from you and Tom so far is just "You have to do this" or "You can't do that" or "You need this" or "You can't just look at that" or "We're not talking about Occam's razor" with no explanation of why you're making the claims you are. It's almost like you're saying "Because I said so."
8:55 AM May 13th
 
studes
You can’t just look at wins as a counting stat, like home runs. Wins occur in a context because the only alternative to a win is a loss. You do one or the other, you HAVE to do one or the other. There are many alternatives to a home run.

Assigning wins to players is an attempt to quantify their total value. If you just stop at wins without adding losses or playing time, you’ve fallen short of capturing total value and you have a stat that just sits there without full meaning. You may like it because it’s simple but we’re not talking about Occam’s razor here. If you want a simple number, calculate win shares above bench, which I used to post at Hardball Times. But don’t settle for just win shares without context.
6:26 AM May 13th
 
MarisFan61
Hopefully Tom comes back and says something more, but as it stands, his entering as he did was very, very strange.
In fact, no matter what else happens from here, it was strange -- because of the nature of what he stated, and even more so, because evidently he thought it was stuff that stands on its own.

(BTW, Tom doesn't reply to my comments, or at least that's been his standing decision, so whatever he says will probably be just in reply to others.)

Strange, because in reply to this article which explains in detail why Steven views the subject as he does and in the context of a Comments section where others are writing similarly, he comes in and gives these cursory pronouncements, without explanation, as though that's just how it is -- factually -- with no controversy or question.
Period.

Tom, is that truly how you see the subject?
(Again, because of what I said above, that's intended only as a figure of speech ["apostrophe," as many of you might know]. I'm not expecting a reply to it from Tom any more than I'd expect one from, say, my baseball mitt after I muff a ball and I ask it why it didn't catch the ball.)

It would be understandable if someone not versed at all in sabermetrics and with a simplistic unsophisticated knowledge of baseball would see the subject as having such a pat answer that is so certain that it doesn't require any elaboration. But actually, of course, it would not be unheard of also for someone extremely well versed in such a field to see his view of it in such a way. Not unheard of at all, but except perhaps in rare instances, unjustified and unfortunate.

Tom, is that truly how you see the subject?

I'm sure I speak for many in saying we hope not.​
11:56 PM May 12th
 
klamb819
Another question: What is the advantage of Loss Shares over (Win Shares)/(playing time)? The advantage is obvious for players with 0 Win Shares, but not otherwise.

And to rephrase Steven's question in a narrower way: What is the advantage of Wins>Replacement over Wins>Zero Value?
4:21 PM May 12th
 
MarisFan61
P.S. ....I'm not talking mainly about why to prefer "WAR" over Win Shares, which might be way too big of a subject (not to mention combative), but at least what he considers to be the reasons for those specifics that he stated.
1:24 PM May 12th
 
MarisFan61
I don't necessarily mind loyalty tests for something like this, as long as the person gives some backup for why we should be loyal to it. :-)

It is odd that such things would be said here without any reasons. The rest of us are saying why we are saying what we are....
1:21 PM May 12th
 
Steven Goldleaf
What is this, Tom, some kind of loyalty test? I use Win Shares for the reasons I explained--I liked Bill's book on it, it made sense to me, etc. But I will give it up and embrace WAR happily if someone with knowledge (like you) just tells me why Wins Above Replacement does the job Wins Share tries to do, but better.

I'm not asking you to write an article, if it's long or complex or subtle or highly detailed, but if that's the case, can you point me to an article or two that go into it?
12:11 PM May 12th
 
MarisFan61
Tango isn't saying "why."

Tom: Why not say the "why"? Why only that-or-that?
11:59 AM May 12th
 
tangotiger
Again, either WAR -- or -- WSh with LSh.

Set aside any desire to use only WSh. That's not a starting point. I can help you with the ride, but only if you set aside your desire in this regard.
11:18 AM May 12th
 
Steven Goldleaf
Tom--what are Wins Above Replacement's advantages over Win Shares? I have no bias in either direction, I use them interchangeably (not within the same study, of course) and view them as equally (approximately) accurate.
10:04 AM May 12th
 
tangotiger
Steven, the better way to think about it is you either want:
- Wins and Losses
- Wins Above Replacement

Those are really your two fundamental starting points. Choose anything else, and insist on it, then it's going to be that much harder to bring you along for the ride.

9:40 AM May 12th
 
Steven Goldleaf
riceman--do you know the story of Bill explaining "Win Shares" to his kid at great length, and the kid thinking the whole time he was saying "chairs" instead of "shares"?

Appreciate all the kind words, folks. I don't want to use these articles as a platform to get in Bill's face, but sometimes I do have an issue that I'm either totally not getting or else just see things differently from Bill on. If Loss Shares is only about providing a context for Win Shares, don't we have all the context (games, IP, PA) that we need? I suspect there's something more, but I'm pretty ignorant on the issues involved in Win and Loss Shares. I've assumed it was something like "Excellent players not only win games but also cause losses in their opponents that we aren't measuring here" (?!?) or "Some players are so bad that they not only have records like 0 win shares but also create negative losses" or something along those lines. My explanations seem to center on marginal events like these, so I'm sure I'm not getting something.
8:31 AM May 12th
 
Riceman1974
By the way, my comment was spoken into my phone, hence the phrase "when chairs" instead of win shares Althoufh in hindsight I kind of like it.
7:32 AM May 12th
 
Riceman1974
I couldn't possibly agree more. I thought when chairs was perfect the way it is. But now, instead of win shares answering the question, "how many games did he win?", it asks the question "what was this players winning percentage?", which opens up a whole Pandora's Box of problems (is a 25-9 player better than a 20-5 player? Beats me.) To me, loss shares hust complicates what was simple. Well said sir.​
7:21 AM May 12th
 
MarisFan61
P.S. to Steven, re Studes' post:
I saw it primarily as an amplification of your point, not a countering of it -- i.e. we don't need loss shares as long as we take playing time into account. But I see that his emphasis could have been either way.
1:23 AM May 12th
 
MarisFan61
I think Steven might be shocked :-) to see me saying that I think he covered it beautifully. Totally beautifully.

Studes notes the importance of taking playing time into account when looking at Win Shares. That's big -- and to me, as long as we know to do that, I really don't understand why Bill ever wanted to revise the system to include Loss Shares. (¿We don't always automatically know the playing time? Sure. Doesn't matter; the win share total tells enough of the story to begin with, then if you want to know more closely about the player or team, take a look at the playing time.) I like the new system a lot, but love the simpler old system -- in part because when loss shares are added, I think it gives the system falsely an impression of being more precise than it is and than it probably can ever be. It rubs me much like how the "WAR" system does by including those decimals. (OK to include them in the sub-totals; silly, pretentious, and somewhat innumerate to include them in the WAR totals.)

I would add, though.....and this is a consideration that I think we don't take into account enough -- I wonder if Bill's newer efforts on the system are somewhat influenced or maybe even primarily determined by his role with an actual team. When he first developed the system, it was a few years before he joined the Red Sox. I think it's very possible that once he got with them and maybe especially as he got more and more into this new life role of his, his approaches became somewhat less theoretical and idealistic and more pragmatic. I can well imagine (although I don't mean it's clear or certain) how a system that includes both wins and losses for each player is more facile for getting the message across to a team's hierarchy.
1:17 AM May 12th
 
klamb819
I couldn't agree more, Steven. And the worst of it is, Bill's pursuit of Loss Shares has no doubt deprived us all of a sortable database of Win Shares.

Studes makes a good point, but there's another solution that's so much simpler: Win Shares per 600 PA. . . and per 1,200 defensive innings. . . and per 160 IP. Maybe those aren't the right numbers; Bill would know better.

But this way, he's holding the best comprehensive Value Metric hostage. Free the Win Share!
9:06 PM May 11th
 
Steven Goldleaf
I don't see it, Studes. "Win Shares needs either Loss Shares or playing time to give it context." I understood from the get-go that 10 WS could either mean an excellent player for a short period or a mediocre player for a long time. Either way, the contribution is about one-third of a MVP-contending year. Saying that it NEEDS context is like saying that 10 HRs or 10 Saves is meaningless because you don't know if they came in a month or in a career. Like I said, every stat is seriously flawed, but Win Shares is far more comprehensible as is. Is this really the crux of Bill's feeling the need to work on developing a Loss Shares system for the past decade or so? I have the sense that there's more here that I'm not getting.
5:38 PM May 11th
 
aagcobb
Loss shares will also let you see who the really bad players are, and that's important as well, or at least fun. If a pitcher gets zero win shares while pitching 140 innings, loss shares can show you just how bad a season he had.
4:51 PM May 11th
 
Gfletch
If we think of Bill James as being baseballs Albert Einstein, then his Win Shares are his Unified Field Theory. I suspect that for both of them, it was and is a search for a chimera, an unrealizable dream.

It's not that the reality doesn't exist. It's that the symbology necessary to explain it can't possibly do so, for the simple reason that the symbology is not the reality.
4:41 PM May 11th
 
studes
Win Shares needs either Loss Shares or playing time to give it context. One way of thinking about Loss Shares is Playing Time - Loss Shares. Without one or the other, you have no way of knowing if a player's 10 Win Shares is good or not.

It's like Pete Rose being the all-time leader in hits. It helps to know that he's also the all-time leader in outs. Hits is simple, Hits + Outs is truer.
4:06 PM May 11th
 
 
©2021 Be Jolly, Inc. All Rights Reserved.|Powered by Sports Info Solutions|Terms & Conditions|Privacy Policy