Problem solved: Wins Above Bloomquist

There’s a war going on in baseball. Did you know that?

I guess in a vague way...kinda like the Afghanistan War.

For single seasons, From 1901 to 2012, (requiring WAR_bat>=-1, WAR_bat<=1 and Qualified for league batting title), sorted by most recent date

Seasons found: 3,262.

For single seasons, From 1901 to 2012, (requiring WAR_pitch<=1, WAR_pitch>=-1 and Qualified for league ERA title), sorted by most recent date

Seasons found: 1,779. 

Bob Ryan is right about this, in the situation where we are comparing players at different positions. You can find a replacement-level CF in the minor league ranks. You can also find a replacement-level 1B in the minor-league ranks. But how can you find someone who is the replacement level for players in general? The stats take away RAR ("Rpos") for being a 1B, and give you extra Rpos for being a CF, indicating that the true replacement-level player is better than the replacement-level 1B, but worse than the replacement-level CF. So the true replacement-level player is the replacement-level 3B? What if he would actually be way worse at CF than the Rpos adjustment would suggest?

The idea of how many runs to add or take away based on position does seem to be an opinion, and cannot factor in the idea that you have a 1B who COULD play 3B (or 2B) but isn't being used there at the moment. No matter what the WAR Accumulation Above All lobby says, that does make him more valuable.

The war on WAR is an unwinnable quagmire.

I'm really surprised and disappointed in Ryan. If you agree with James' stuff, there's absolutely no reason you should be saying that WAR is a nonsensical concept. (You could of course argue about how to figure it out, but not that it's an illogical concept.)

If Ryan had acted like a reporter and given James a call rather than just speculating about James' feelings, I'm sure Bill would have told him in so many words "it's not a stat I personally use, but it makes perfect sense."

Well, is Ryan not correct in that there is still a lot of debate about how to define replacement?

Also, by the time you get any kind of reliable baseline about how good a player is with the glove, in many cases, he's no longer at that level anyway, as it is my understanding that the defensive peak is ordinarily reached a few seasons before the batting peak.

If Ryan had acted like a reporter and given James a call rather than just speculating about James' feelings, I'm sure Bill would have told him in so many words "it's not a stat I personally use, but it makes perfect sense."

I wouldn't be so sure about that. James has said that he doesn't consider WAR to be sabermetrics at all.

Well, is Ryan not correct in that there is still a lot of debate about how to define replacement?

Sure. Does that make WAR "complete nonsense?" Of course not. Another talking head overstates his case, yipee.

I come from that quaint generation of yesteryear

Could not have said it better myself.

I thought it was wins above Scott Cousins.

when i evaluate players i prefer to center around the league average because you get a hard number baseline that way and the judgment call is taken out of it. however it does punish defensive specialists and pitchers by not giving them credit over replacement level for hitting. an ops+ of 78 then becomes a big negative rather than just a smaller positive, and one that defense and baserunning can't bring back to positive territory even though perception of these players was that they were above average and benefited the team above an average team. this could mean i need big positional adjustments, or it means our perception of their value is distorted. i'm not sure.

however, when i've evaluated players, i was surprised to find that defensive dropoffs were by far the most reliable aging indicator. psychologically, i'd think that a great defender would remain "savvy" in the field while his bat declined, but no, it is the opposite. defensive declines seem to be by far the most universal, consistent, and reliable. they lose their range before their speed on the basebaths or their bat.

Every stat we have includes some judgement.

when i evaluate players i prefer to center around the league average because you get a hard number baseline that way and the judgment call is taken out of it.

I suppose this can work if you're evaluating MVP candidates. The problem is that league average players, or just below league average players still have value and are scarce.

While writing this, my daughter mentioned a garage band she knows called Above Average. One of her friends says the band is really below average. I suggest perhaps they should lower their bar and call themselves Above Replacement Level.

i guess i continue to struggle why anyone would be oppposed to another means of measuring success in baseball.

and the author needs to stay in closer touch with things. i am just guessing here but i suspect that if you polled those aware of bill james they would tell the author they think james is kind of crank these days. which is not a poor assessment

While writing this, my daughter mentioned a garage band she knows called Above Average. One of her friends says the band is really below average. I suggest perhaps they should lower their bar and call themselves Above Replacement Level.

Or the 'Ments, for short.

Every stat we have includes some judgement.

What Darren says is absolutely true. If you're going to write off every stat that includes judgement, you'd better throw out Batting Average and ERA, Bob.

I wouldn't be so sure about that. James has said that he doesn't consider WAR to be sabermetrics at all.

If this is true, it says more about James' current state than it says about WAR (to echo Harveys in post 15 in different words).

Which stat has the least amount of judgment? Perhaps intentional walks. Even plate appearances are corrupted by the pernicious phenomenon of the catcher's interference.

when i evaluate players i prefer to center around the league average because you get a hard number baseline that way and the judgment call is taken out of it.

The problem with average as a baseline, is that it punishes below average players for playing time, even though their contributions are valuable. Would you rather have a starting pitcher who gives you 220 innings of 4.20 ERA, or one who gives you 150 of the same?

And replacement level is just league average minus about 2 wins per full season of PA/IP. You can argue about the exact level replacement should be at. But it's it's pretty much inarguable that it gives you a vastly more accurate description of actual value, than average does. And since both are based on the exact same "hard number"...

Which stat has the least amount of judgment? Perhaps intentional walks. Even plate appearances are corrupted by the pernicious phenomenon of the catcher's interference.

Are you kidding? Who are you to judge a pitcher's intent?

The least judgmental stat can't even be Games Played, since there's the Larry Yount problem.

News Flash: Measuring players by WAR is imperfect.

News Update: Just like all other statistical measures of players, just less so.

Which stat has the least amount of judgment?

A Run Scored is near-absolute: only the guy who actually touches the plate gets credit. But of course, even there, there's a conventional scoring decision to credit the run to the guy whose feet touched the base, even if he hit an easy ground ball and was safe on a force-out, (1 AB, 0 H), while the guy who got a previous hit or walk is out and given no part of the Run.

So oddly enough, scorers sometimes give someone an RBI for making an out, and also sometimes give someone a R for making an out. The RBI out is sometimes arguably "productive"; the force-out followed by a score is a dead loss overall.

Are you kidding? Who are you to judge a pitcher's intent?

I believe they are only awarded when the catcher is doing his standing and pointing thing, that is to say, when intent is beyond dispute. I suppose it's possible that once in a century a pitcher will try and sneak a 3-0 fastball down the middle on the unsuspecting hitter, but miss wide outside.

Games? As in appearances for pitchers?

I believe they are only awarded when the catcher is doing his standing and pointing thing, that is to say, when intent is beyond dispute.

Perhaps the scouting reports said that, for example, Barry Bonds was especially vulnerable to pitches shoulder-high and two feet outside.

While writing this, my daughter mentioned a garage band she knows called Above Average. One of her friends says the band is really below average. I suggest perhaps they should lower their bar and call themselves Above Replacement Level.

If they aren't above average, they certainly aren't above the level of The Replacements.

Sunday, November 18, 2012

Bill James Online Mailbag [...]

Dear Mr James,

I’m a student at DePaul University writing a news story on what I’m calling “The War on WAR,” essentially a look into the different methods(BB Prospectus, Fangraphs, and BB reference) of calculating Wins Above Replacement. In light of this years AL MVP debate the WAR statistic is getting more publicity than usual and since there is not a standardized way of getting a true WAR of a player I was wondering how you personally view the different methods of calculating it. Thanks,

No; I don’t use that statistic myself, so I don’t have any view of how it should be calculated.  Not to speak ill of it; it is a useful concept.  It is probably in everyone’s best interests that there be multiple ways of calculating it.  It is like, for example, the Money Supply in the Economy.  There are all kinds of different ways of calculating the supply of money in the economy, or the GDP (Grounded Into Domestic Products.)  There are many different ways of estimating the GDP, and there SHOULD be, because if something is real, it can be estimated in different ways, and it weakens the value of the stat to reduce the number of ways that it can be calcualted.  [Emphasis added]

http://www.baseballthinkfactory.org/newsstand/discussion/bill_james_online_mailbag4/

Tango on Bill James and WAR

http://www.insidethebook.com/ee/index.php/site/comments/bill_james_turns_win_shares_into_wins_above_replacement_war/

To convert Win Shares to Wins Above Replacement, you do: WS/3 - .333 * (WS+LS)/3That “/3” is to convert win shares into wins.  And the “.333” is the win percentage replacement level.  So, the final simple equation is (2*WS-LS)/9.  For McGriff, that’s 319*2 - 152 or 486.  And divide by 9 to get to 54.  This is how you convert Win Shares and Loss Shares into Wins Above Replacement (WAR).  And Bill James has ranked his firstbasemen by (his version) of WAR. [...]

It seems to me therefore that Bill has accepted that some sort of wins above baseline is needed.  And he is using a level that pretty much match what we’re using for wins above replacement, since the results are consistent.  I wouldn’t worry about the implication of the “.333 win percentage”, since that number is specific for the way his system is constructed.

This really brings Bill James back to where he always was 25 years ago, when he first wrote about replacment level in the context of Rice v Guidry and Clemens v Mattingly.  And when he ranked players as “chance of being better than a .400 player”.  Bill took a detour along the way in terms of ranking players by Win Shares (without considering Loss Shares).  Him bring Loss Shares into the mix really resets his position back along the Wins above replacement track that the rest of us are on. [...]

Ok, I can buy he’s using .350, not .333. The main point is that Bill is ranking based on some form of wins above replacement.

And, wins above replacement is nothing more than Linear Weights at its core.

Which stat has the least amount of judgment?

A Run Scored is near-absolute: only the guy who actually touches the plate gets credit. But of course, even there, there's a conventional scoring decision to credit the run to the guy whose feet touched the base, even if he hit an easy ground ball and was safe on a force-out, (1 AB, 0 H), while the guy who got a previous hit or walk is out and given no part of the Run.

I was thinking of Runs as well, but I suppose sometimes there is a tough call at the plate and it's either a run or an out, depending on the umpire's judgment.

Wellll... I don't think Mr. Ryan is THAT far off on some things. Sure, there's an element of "judgement" in all stats (if you broaden "judgement" to include scorer's decisions; the manager's "judgement" in ordering IBBs, defensive shifts, calling for sacrifices, lineups, etc.; the centerfielder's judgement --or lack thereof-- on balls hit over his head; the baserunner's judgement in trying to score from first on a ball hit to the gap), but the traditional stats rely on basic arithmetic MUCH more than WAR. I think it's a useful stat (any stat that reinforces the true greatness of, say, Mickey Mantle while cutting through the nostalgic haze that somehow suggest Jack Morris is a HOFer, is useful), but we're a long way off from refining it to the point where it should be accepted as the uberstat. There are just too many weird things that pop up (forget Miggy vs Trout -- I'm in the Trout camp). I'm talking about things like huge swings in defensive dWAR from season to season, etc).

The good news is that, at least to my untrained eye, league leaders in WAR generally align with leaders in more traditional stats. Top-5 in OBP + slg will generally get you top-5 in WAR (unless the guy is a butcher in the field, or has a huge home/rd split). But WAR is still very much in the "evolving" stage.

There are just too many weird things that pop up (forget Miggy vs Trout -- I'm in the Trout camp). I'm talking about things like huge swings in defensive dWAR from season to season, etc

Why is that weird, or wrong? We accept huge swings in offense from season to season as normal. Why wouldn't you expect large variations on the defensive side of things?

Even games -- in fact even games started -- relies on judgment. Specifically (at the very least) the judgment of the umpires that it is not raining too hard to continue to the fifth.

I wonder if Ryan knows that the replacement-component benefits Cabrera more than it does Trout or if he just assumes that all parts of the stat are biased in favor of Mike.

It’s how much more Player X is worth than a player that doesn’t exist!

If it bothers you so much, use WAA. Trout was 8.8 wins above average, Miggy 4.7 It's such a cop out for these knuckleheads to throw their hands up and say "Since we don't really know what a replacement level player is, we can't really know if Trout was better than Cabrera." Yes you can. You know what an average player is, and Trout was twice as better as him than Cabrera was. Problem solved.

Or not. These guys aren't looking for information to base an opinion on. They use information like a drunk uses a lamppost: For support rather than illumination.

Which stat has the least amount of judgment?

Home runs. Yeah, there's the odd fair/foul call, and the even rarer "did it hit over the line or under?", but as a percentage of the whole, the vast majority of HRs are judgement free.

I wonder if Ryan knows that the replacement-component benefits Cabrera more than it does Trout or if he just assumes that all parts of the stat are biased in favor of Mike.

Why is that, anyway?

The replacement-component adjustment is just a playing-time (and league) adjustment.

Cabrera got 697 PA, so to get from RAA to RAR they add (697/650)*22, which is 23.59. His RAA was 45 and his RAR was 69.

Trout got 639 PA, so to get from RAA to RAR they add (639/659)*22, which is 21.63. His RAA was 85 and his RAR was 106.

It looks like Trout's number was rounded down to 21, and Cabrera's number was rounded up to 24. But I assume that's misleading and the real numbers were more precise than the 2 significant figures displayed on BB-ref.

Home runs. Yeah, there's the odd fair/foul call, and the even rarer "did it hit over the line or under?"

Then there is the rarest of all calls... http://mlb.mlb.com/video/play.jsp?content_id=3977861&topic;_id=&c_id=mlb&tcid=vpp_copy_3977861&v=3

Wins above 175% the player Emmanuel Burriss is.

Was this covered already? Anyway, the choice of replacement level has no impact on the difference between two players (#1 and #2 below):

R1 = X1 - R
R2 = X2 - R

Where X1 is the raw "runs" a player is credited with and R is the level a replacement player is credited with ... making R1 the runs above replacement (which can then be converted to wins).

R1 - R2 = X1 - X2 - R + R = X1 - X2

The value chosen for R doesn't matter in measuring differences. You can use league average if you want.

bWAR and fWAR disagree in terms of RAR for a given player in part because of choosing different replacement levels. In the end, that's such an unimportant difference they should simply agree on what replacement level is. But to the extent that bWAR and fWAR differ in terms of a gap between two players (say MT and MC) then they differ in substantive ways about how best to calculate X. That's a topic for serious discussion.

Of course the only alternative to that reality is the dreaded groupthink where all the projection systems agree completely and that uniformity would form the basis for criticism of the methodology.

It looks like Trout's number was rounded down to 21, and Cabrera's number was rounded up to 24. But I assume that's misleading and the real numbers were more precise than the 2 significant figures displayed on BB-ref.

I think there's a slight adjustment based on batting order position, otherwise leadoff hitters get replacement value just due to more PA. Technically Rrep might be "replacement level player with the same mix of batting order positions."

Trout got 639 PA, so to get from RAA to RAR they add (639/659)*22, which is 21.63. His RAA was 85 and his RAR was 106.

You made at least one typo here. I assume you mean 650 as the denominator for Trout since that comes out to 21.63.

Given the rounding, it doesn't matter a lot. Trout rounds to 21 for everything from 654 to 686 as his replacement PA. It's just tinkering down in the first decimal place.

There’s a war going on in baseball. Did you know that?

Actually, the war is over. Your side lost. Quite some time ago, in fact.

I wish I were making this up. I’m not. Here is the explanation for WAR/WARP on page xiii of the 2013 “Baseball Prospectus”:

That's the explanation for WARP. WAR and WARP are different things.

It’s how much more Player X is worth than a player that doesn’t exist!

Replacement level players exist everywhere; they are ubiquitous, which is the point. They are as freely available as house painters.

It’s ultimately based on a judgment. It’s not a statistic!

Where to set replacement level is certainly an issue. But I don't see why that invalidates the stat.

Was this covered already? Anyway, the choice of replacement level has no impact on the difference between two players (#1 and #2 below):

R1 = X1 - R
R2 = X2 - R

Where X1 is the raw "runs" a player is credited with and R is the level a replacement player is credited with ... making R1 the runs above replacement (which can then be converted to wins).

R1 - R2 = X1 - X2 - R + R = X1 - X2

The value chosen for R doesn't matter in measuring differences. You can use league average if you want.

That's only if R is constant, which it isn't, since it is dependant on playing time.

The war on WAR

Mr. Ryan: you, sir, are not my typewriter!

The main point is that Bill is ranking based on some form of wins above replacement.

And, wins above replacement is nothing more than Linear Weights at its core.

The irony here was lost on most readers back when Win Shares came out a decade ago (has it really been that long?). James published a number of articles in the 1980s about how senseless it was to come out with The Ultimate Baseball Statistic. Then, suddenly, when his relevance was at an all-time low, he came out with his own version of WAR.

I'm kind of a strange guy. I've had this project going on for a few years now, where I've been rereading all the classic Sabermetric books I can get my hands on in order. I've got the Abstracts from 1982-1988 (even the reprints of 1977-1981 are almost impossible to find), as well as both Historical Abstracts, The Hidden Game of Baseball, Win Shares (and a bunch of other random James books), The Book (first edition, I believe), every issue of The Hardball Times (but not much of a Prospectus collection -- probably because I've always visited Primer / BBTF), some random old issues of The Baseball Research Journal, and an almost complete collection of The APBA Journal from the late 1970s to 1993 or so (I think I've got some old Strat newsletters sitting around as well).

Anyway, I'm almost done with The Hidden Game of Baseball. Though I don't have access to everything James had published up to that time, I've got to say that linear weights blows a lot of the stuff James was doing out of the water. The statistic is easy to use, the reasoning for each calculation is clearly explained, and you don't see any of the half-baked, rushed ideas that James frequently produced.

The entire concept behind Win Shares contradicts a number of things James was saying again and again in the 1980s, in what I believe was an attempt to stop Palmer from stealing his thunder. Honestly, though, I firmly believe that linear weights and The Hidden Game of Baseball have had a much larger impact on the development of sabermetrics and baseball statistical analysis than runs created and the Abstracts.

I still disagree with a number of Palmer's ideas. I find irritating the idea that strikeouts are overrated, for example, or that BAA doesn't tell us as much as how many earned runs a pitcher gave up. I also haven't taken the time to deconstruct the way Palmer calculated park factors, since I figure that ship sailed about 30 years ago or so.

Still, though, linear weights, WAR and the concept of removing biases uniformly in baseball statistics are very powerful. Reading that one book has done more for my understanding of current sabermetrics and why we measure the things we do (things like win probability) more than anything I've read by James.

"I'm almost done with The Hidden Game of Baseball."

I have mentioned several times on this site that I have a galley proof of this book from that era (lime green cover), and it's not so I can engage in a bidding war. I would sooner meet people who wanted it, and if they can afford any sum at all, we can give the proceeds to charity. Or maybe the most sincere applicant gets it for free, and pays it forward.

Never an answer, which costs me nothing. It doesn't mean much to me, but I always wondered if others here would diagree. More bemused than anything else; how many galley proofs are actually out there?

it's not a stat I personally use, but it makes perfect sense.

Win shares is absolutely a WAR.

#45 Linear weights blows away runs created if and only if you use the "slope correctors" As Bill pointed out, if he used a slightly different formula for runs created in every year he could totally get the standard error of runs created down.

Baseball-reference uses a linear weights approach with no slope corrector. But they calculate the value for the events for every season.

The most accurate linear weights approach with no (as my friend Bob Rich called it "cheating") individual season adjustments is our own Jim Furtado's extrapolated runs. (and it is indeed better than runs created, though it does have some method errors. In particular, counting DPs without adjusting for context)

That's only if R is constant, which it isn't, since it is dependant on playing time.

True but nobody ever cares about comparing part-timers and full-timers.

I do respect the New Math of baseball.

WAR is nonsense.

I'm not really getting respect there. I feel like he's playing at acting like he's trying to understand it, when in reality a 30-minute discussion with any number of people would answer most, if not all, of the questions he's raised.

Someone should come up with a list of 30-40 players at each position who grade out at "replacement level". Not just current guys, but players from the 70s and 80s. Give it a Potter Stewart feeling - you may not know how to calculate replacement level, but you know it when you see it.