I'm a bit dim, when it comes to maths, so I found Tango's thoughts on all this helpful in deciding whether this is an 'upgrade', as it is promoted, or a 'change', my sceptical first reaction to the headline alone.

First thing I did when I saw this was check Zobrist's page because the old B-R WAR didn't like his defense at all in 2011. With the new one he now is at 8.4!, second in the AL for that season by .1 behind Verlander. That's a big change there, not sure what the number was previously but I know it was dramatically lower.

Dear Mr. Forman:

Your new formula claims that Jose Bautista month's worth of games have been exactly as valuable to the Toronto Blue Jays as Omar Vizquel's 12 PAs of burnt toast production (.091/.161/.091). Please correct this glaring error in the formula before I throw myself off the nearest bridge, as it is simply impossible that such a thing could ever happen in this world.

Yours,

Every single Blue Jays fan.

Wow, the changes in the method make for some dramatic re-evaluations of the players. Roberto Clemente, for example, looks far better than he's previously been given credit for.

The new WAR seems very pitcher-friendly. In this system, 29 of the last 50 NL MVPs are pitchers, 27 of the last 50 in the AL.

The new WAR seems very pitcher-friendly. In this system, 29 of the last 50 NL MVPs are pitchers, 27 of the last 50 in the AL.

So another reason to ignore war for pitchers? Been doing it for a while, apparently no reason to stop.

So another reason to ignore war for pitchers? Been doing it for a while, apparently no reason to stop.

Depends, I know a lot of people who honestly believe that a great pitcher is worth more than a great position player, can't say I can prove 'em wrong.

Depends, I know a lot of people who honestly believe that a great pitcher is worth more than a great position player, can't say I can prove 'em wrong.

70 years ago, sure. 40 years ago, maybe, post 1980, not even remotely close.

You would have to fully surrender yourself to the Dips alter, and fangraphs war to really believe that. Yes occassionally a great pitcher will be worth what a great position player is worth, but those are going to be exceptional years, not the norm.

There is a new best player not in the Hall of Fame. And his name is...

...Jim McCormick. A 19th century pitcher with a record of of 265-214. His 83.2 WAR is right between George Brett and Robin Roberts.

70 years ago, sure. 40 years ago, maybe, post 1980, not even remotely close.

See, I would think increased run environments would make great pitching scarcer, so getting hold of one of those top 5 to 10 guys in the league would make them even more valuable. Its not the Dips alter you have to sacrifice to, its the positional scarcity alter. Personally I agree with you that an everyday player is more valuable, but I don't think the argument the other way is incoherent or anything.

We are going to need a rewrite on a few of those bullet points, in that I can actually understand what a couple of them are trying to say. Let's get on that.

The last bullet point seems to be Sean just ####### with us. So oWAR didn't change but you change it's name, but dWAR did change but you left it's name the same. Got it. I guess.

Can someone summarize my new opinions on the Hall worthiness of notable players? Thanks.

Personally I agree with you that an everyday player is more valuable, but I don't think the argument the other way is incoherent or anything.

I don't know, it's hard to quantify. I've actually been thinking about stuff like that the last few days with Longoria out. The Rays can lose him, their best position player, for a month and not see a significant change in their expected win total even if they ran slightly above replacement level Elliot Johnson out there every day. What would happen though if Alex Cobb was eaten by a shark tomorrow and then Shields or Price had their head smashed by a line drive in their next start? Other than the possibility of moving Davis back to the rotation they'd be looking at a pretty bad replacement and going from a starter that might go 7-8 innings allowing 2-3 runs to one that will often go 5-6 innings giving up 4-5 runs is a damned big difference both in the chance of winning each of those games and also the strain it puts on the bullpen.

Injury risk and the greater unpredictability in general of pitcher performance prevents teams from really voting with their wallets on the issue but I suspect that if they felt that pitchers were as reliable as position players to stay healthy and good then the highest paid players in the game would be a bunch of starting pitchers.

Also, the offense WAR and defense WAR don't add up to a player's total WAR, so maybe some more explanation there?

Pitchers really did seem to receive a boost. I hope I'm not falling victim to the "I'm not used to it so I don't like it" idea, but so far, I don't like it.

Folding ROE, infield singles, SO vs. Non-SO into wRAA.

Not sure why you'd fold ROE into it but not Rdp. Both are largely a function of handedness and speed. Rdp does have an opportunity component to it that's unrelated to batter performance but that's a pretty thin reed. Anyway, doesn't matter much. Also, I liked having rROE separated out.

Excluding pitchers’ hitting and averaging by league rather than year from the league averages for wOBA and wRAA.

Excluding pitchers would seem to make sense for position player WAR but doesn't it screw with total WAR?

But more importantly, doesn't the league-averaging thing mean I can't do a straight comparison of players across leagues?

Use of a player-influenced runs to win conversion for both batters and pitchers based on PythagenPat.

Would somebody like to translate?

I’ve changed the oWAR and dWAR formulations. oWAR is now called ndWAR for (no-defense WAR), but is the same otherwise. dWAR now contains the position component of WAR, so the Career Leaderboard is now dominated by SS, C and other great defensive players.

I'm confused. Didn't oWAR used to have the position component, meaning it was offense relative to position? If that's now in dWAR (which is OK but has its downsides), then there has been a change to oWAR. Or is the position adjustment now in both which means WAR ne ndWAR + dWAR?

OK there is in fact no ndWAR but rather ndRAR. The old oWAR appears to be (as of 5 seconds ago at least) afWAR (with no idea what af stands for) but does still include the position adjustment. So the help tag explicitly states WAR ne afWAR + dWAR ... which seems an odd choice to make.

Am I the only person who thinks changing the way these stats are figured hurts Sabremetrics growing foothold in the public's eye?

The basement community has been beating these advanced metrics down our throats for a few years now. Stating that such-and-such player was robbed of an MVP or Cy Young, or that so-and-so didn't deserve theirs. All based on these foolproof numbers. Sure they say WAR is imperfect, but not by much.

And now, the mathletes go through and change the formulas? Like mentioned above, now Zobrist was the most valuable position player in 2011. I don't recall his name being in the MVP chase last year, but I bet it would have been under this new formula.

To me, this puts a bit of a black cloud over the Sabremetric community. Like the numbers can be skewed to mean whatever you want them to mean (which is true, but we are all hoping for objectivity). How are we supposed to take solace in the numbers, and trust them, when they can change? It's like re-writing history.

I dunno, dWAR has always been ########.

Like mentioned above, now Zobrist was the most valuable position player in 2011. I don't recall his name being in the MVP chase last year, but I bet it would have been under this new formula.

I'm pretty skeptical about the defensive credit that Zobrist is getting for 2009 and 2011. While it was obvious that the previous defensive system was dead wrong about him (he was rated as below average at 2B in 2011) I don't know if a second baseman or right fielder is going to be able to save as many runs as he's getting credited with. Here are the rtot/yr (approximately 1200 innings) for him along with his actual innings played:

2009 2B: 27 (714.2)
2009 RF: 33 (329.1)
2011 2B: 26 (1058.1)
2011 RF: 25 (289.2)

Small sample sizes for the RF numbers of course but in both seasons the 2B numbers are in the upper 20s, that just seems too high. Zobrist is a great player and his versatility/the Rays roster construction leads to him being unfairly left out of the discussion about who the best 2B is because he plays RF a fair amount but according to new WAR he was the most valuable position player in the AL two of the last three seasons and that doesn't pass the smell test. Nearly 30 runs saved per 1200 innings for a 2B is just excessive.

This doesn't even come close to passing the smell test. Matt Kemp the 9th best player in MLB in 2011 and the seventh-best this year? Are you friggin' kidding me?

Cool, all the old threads can be re-argued! Shall we start with Ichiro or Morris?

The basement community has been beating these advanced metrics down our throats for a few years now.

A sense a fully unbiased and objective point is about to be made.

Use of a player-influenced runs to win conversion for both batters and pitchers based on PythagenPat.

Would somebody like to translate?

We actually had a very nice little discussion about runs per win a few days ago. I read up a bit on it then, unfortunately, by the time I had, the discussion pretty much died.

It's basically a way of translating runs into wins, rather than saying something silly like 10 runs equals 1 win, which may or may not be reasonably accurate depending on the run environment. The Pythagorean records we all know and love are derived from it.

Anyway, unless I messed up somewhere, the formula should be: RPW = (RPG^(1-z))*2 where RPG is for both teams, and z typically is 0.287.

Also, there is a cool graph from tango, showing RPW over time here.

This doesn't even come close to passing the smell test. Matt Kemp the 9th best player in MLB in 2011 and the seventh-best this year? Are you friggin' kidding me?

That does seem rather nonsenseful.

Dwight Gooden now has the best post-deadball season in the history of baseball. By far.

Zach Grienke has the best season of the last 10 years.

Dwight Gooden now has the best post-deadball season in the history of baseball. By far.

Zach Grienke has the best season of the last 10 years.

And yet, Dennis Eckersley's 1992 still underwhelms.

How are we supposed to take solace in the numbers, and trust them, when they can change?

The quick answer is that these numbers are always an interpretive estimate of what players contribute to winning. They're there to be criticized, not for "solace." And if somebody comes up with a better estimate, why shouldn't they abandon the old one?

The only people who should be put about by an upgrade in WAR formulas, as several posters above have noted with telegraphic snark, are those who had become addicted to looking at the single number in the WAR column and ending thought at that moment.

In 1998, Kenny Rogers, Chuck Finley and Al Leiter were all more valuable than both Mark McGwire and Sammy Sosa.

Like the numbers can be skewed to mean whatever you want them to mean (which is true, but we are all hoping for objectivity).

I know I'm considered a lone nut for saying these kinds of things, but there's no way that WAR numbers should ever have been considered truly objective. Or ever will be.

In 1998, Kenny Rogers, Chuck Finley and Al Leiter were all more valuable than both Mark McGwire and Sammy Sosa

Mind you, I didn't say the new numbers were a better estimate. I will have to study the issue of whether Rick Reuschel's 1977 was more valuable than Willie Mays's 1962 before deciding :)

What could do real good for sabremetrics would be published error bars. And no false precision by using more digits than what is significant.

Matt Kemp the 9th best player in MLB in 2011 and the seventh-best this year?

What.

And no false precision by using more digits than what is significant.

The problem is that while 0.1 WAR is clearly insignificant, 0.9 WAR, while not necessarily conclusive, is fairly significant.

Not compared to 1 WAR.

Cespedes fielding after 26 games:

TZ: 0
UZR: -3
DRS: -11(!)

He's been bad, but is that even possible?

What could do real good for sabremetrics would be published error bars. And no false precision by using more digits than what is significant.

This would only give more fodder to the people who reject it out of hand.

The problem is that while 0.1 WAR is clearly insignificant, 0.9 WAR, while not necessarily conclusive, is fairly significant.

How is that a problem? 0.9 to one significant digit is 1.

Use of a player-influenced runs to win conversion for both batters and pitchers based on PythagenPat.

Would somebody like to translate?

A player's performance influences his run environment, which needs to be accounted for when making run-to-win conversions. A great pitcher can turn a 9-run environment into a 7-run environment, or a great hitter can turn it into a 9.5-run environment, and that is now accounted for...


...incorrectly, it would seem. The Book blog discussion turned toward the proper way to make these calculations, because the factors for pitchers were too small, giving them too much credit. Pitchers' numbers will stop looking so crazy once this gets ironed out.

Cespedes fielding after 26 games:

TZ: 0
UZR: -3
DRS: -11(!)

He's been bad, but is that even possible?

That was my response when I saw that Brett Lawrie (94 OPS+) was tied for the AL lead in WAR for position players because he's +12 for fielding through 27 games. I suppose it's possible, but he's not even in the top 25 on Fangraphs.

The problem is that while 0.1 WAR is clearly insignificant, 0.9 WAR, while not necessarily conclusive, is fairly significant.

How is that a problem? 0.9 to one significant digit is 1.

If Player A has 0.5 WAR, and Player B has 1.4 WAR, then that is information I want to have. I don't want to see two 1's.

And the reality is that the difference between the two might be something like 6 to 8 runs and over a whole season that is essentially meaningless. Or hell, if you want the numbers could simply be 1, 1.5, 2, 2.5, 3, so on and so on. Or we could just turn each when into a slice of a win. Perhaps 3 slices to a win and we'll call each slice a "share".

And the reality is that the difference between the two might be something like 6 to 8 runs and over a whole season that is essentially meaningless.

I assume by 'meaningless', you mean 'enough to get the Red Sox into the playoffs last year', and 'enough to prevent the Cardinals from winning the WS last season'.

I'll try to respond more later, but it is pretty clear that the manner in which we are converting runs to wins in breaking down for extreme pitching performances. I implemented a system that has been proposed by Tango and used by FanGraphs (though I can't recreate the numbers they are showing). I think the issue is that when you have 2009 Greinke on the mound we are treating it like a league where team's average 3.50 runs per game (rather than 4.50) and in the first league you only need like 7-8 runs per win compared to 10 for the second.

We are discussing details on Tom's board, but it looks like a better approach would be to look at the equations for a case where a 2.50 team (with Greinke) faces a 4.50 team (average) and work out the changes in wins there. It's quite a bit more complicated, but I think it will handle the dominant pitcher performances much better and not penalize the dominant hitters as much either.

I apologize for not foreseeing this issue. I was a bit unsettled by the war_rate's we were generating, but I'd seen multiple signoffs on it, and thought it was the correct path.

Here are the runs above replacement for last year. The reason Kemp drops is because of the runs to wins values. As for Kershaw and Halladay, that is due to quality of opp, park factors and Phillies defense is considered much, much worse than the Dodgers.


Greinke a

That was my response when I saw that Brett Lawrie (94 OPS+) was tied for the AL lead in WAR for position players because he's +12 for fielding through 27 games. I suppose it's possible, but he's not even in the top 25 on Fangraphs.

Starlin Castro's at the top of the NL list, despite being on pace for about 45 errors this year.

EDIT: As if on cue, Castro just made another error.

I assume by 'meaningless', you mean 'enough to get the Red Sox into the playoffs last year', and 'enough to prevent the Cardinals from winning the WS last season'.

No, I mean meaningless by being well within the margin of error.

No, I mean meaningless by being well within the margin of error.

When you say '9 runs is in reality 6 runs, which is meaningless', I don't see how I am supposed to interpret that as '9 runs is within the margin of error'.

Either way though, I stated initially that 9 runs is not conclusive, which is the takeaway from something being within the margin of error. There is however a vast gulf between 'not conclusive' and 'meaningless'.

When you say '9 runs is in reality 6 runs, which is meaningless', I don't see how I am supposed to interpret that as '9 runs is within the margin of error'.

.9 WAR doesn't necessarily mean 9 runs.

Either way though, I stated initially that 9 runs is not conclusive, which is the takeaway from something being within the margin of error. There is however a vast gulf between 'not conclusive' and 'meaningless'.


I understand now, it's pedantry.

.9 WAR doesn't necessarily mean 9 runs.

I understand now, it's pedantry.

Wheeeeeeee!

Brad Radke’s 2000—in which he went 12 and 16 with a 4.45 era—was more valuable than Sammy Sosa’s 1998 (in which he hit 308/377/647 with 66 homers). (6.8 to 5.9 WAR)

Prince Fielder’s 50 home run season in 2007 (on a 288/395/618 line) ranks as less valuable than Omar Olivares’s 1996 effort, where he went 7 and 11 with a 4.89 era in 160 innings for the last-place Tigers. (3.6 to 3.8 WAR)

1989, National League:

Lonnie Smith 8.3
Will Clark 8.0

That's certainly not what it used to say. In fact, it used to have Clark way out in front at 9.4. Can anyone explain what happened there?

This is going to take some studying and getting used to. There have been some big changes for some players when it comes to dWAR. See Yaz, Al Kaline, Sam Crawford (Joe Jackson, too, in a much shorter career), Frank Robinson, and Keith Hernandez, just for starters, for players who have nosedived (Crawford wasn’t thought highly of by the old system anyway, but now he’s in the nether regions). Tris Speaker wasn't thought that highly of under the old system, either, but he's fallen further under the new one. Nothing like the way Tyrus Raymond has, though.

Stan Musial and Hank Aaron and Albert Pujols, too, have suffered a great deal in their defensive rating. Babe Ruth, too, was viewed better under the old system, as was Barry Bonds. Barry went down, but is still positive; Ruth is now a negative. Williams and Mantle under the old system were slightly negative, I think; now they’re both big negatives in dWAR.

Besides formerly great fielders falling, like Yaz and Kaline and Hernandez, some who were valued highly for their defense have rocketed up even more, though. Mays, Brooks Robinson, Ozzie Smith, and Mark Belanger (Belanger now has as much aggregated dWAR as Smith in a much shorter career—and he was a tremendous fielder, all his contemporaries agree, I believe), with a lot of dWAR under the old system, have gained even more credit on defense under the new system. So has the redoubtable Cal Ripken, who came out very well under the old system, but he is now in the realm of the super great fielders. And he very well might be one, but the difference between him and Jeter in fielding is almost other-worldly.

We had some sort of discussion about Hornsby a few months ago, and some of us were surprised to see that WAR had his defense as better than Collins or Morgan then. Hornsby's defensive rating, in the teeth of the conventional wisdom of his time, was high under the old fielding evaluation--now, he's a great fielder, it looks like. See also Grich and Trammell. Honus at 21.7 I think went up, but no matter what the system, this guy never suffers. This will all take some learning and explaining, I think.

dWAR is completely different from what it was before. It is now fielding + position. Before it was just fielding.

If you want to compare two fielders you never should have used dWAR, you should be using runs from DRS or TZR, rather than messing with the conversion to wins.

I love pretty much everything Sean does but this has all the makings of a disaster.

dWAR is completely different from what it was before. It is now fielding + position. Before it was just fielding.

Is overall league-wide dWAR still negative?

I love pretty much everything Sean does but this has all the makings of a disaster.

I haven't studied it yet, so I don't know if it actually IS a disaster. But clearly it's going to be perceived as one.

Some strange results to be sure, and a lot of them. Thats never a good sign.

I didn't mean to suggest in my post above that I thought the new WAR was a disaster or anything like that. At this point, I don't know enough about it to make such a definitive judgment. I just was pointing out some changes that stuck out to this mind.

Boy, I tell you, though--how great was Brooks Robinson on defense? Most everyone after him on the leader board for a good ways are shortstops. He provided more dWAR than the great shortstops!I always thought Nettles and Clete were fairly close. How great was that Oriole infield when you had Grich at second to go along with Belanger and Brooks?

There will be some interesting discussion that will arise out of all this. That can't be anything but good.

dWAR is completely different from what it was before. It is now fielding + position. Before it was just fielding.

I hate this. Keith Hernandez, Will Clark, JT Snow, Jeff Bagwell, Mark Grace, Don Mattingly, Wes Parker ... they all have negative dWARs. I realize that this all might make sense if you understand what the statistic really means, but come on. This statistic needs to be monkeyed with if anyone is going to trust it.

How great was that Oriole infield when you had Grich at second to go along with Belanger and Brooks?

I can't speak for the recently updated version of WAR, but I know under B-ref's recently outdated system, the 1973 Orioles rated as the best fielding unit ever, with +11.2 dWAR.

I do appreciate that instead of platitudes about how things are always getting better and will continue to get better, Sean's explanations of this amount to "dWAR became too popular, I had to change it to make less intuitive sense so people would stop using it inappropriately".

Heh.

I can't speak for the recently updated version of WAR, but I know under B-ref's recently outdated system, the 1973 Orioles rated as the best fielding unit ever, with +11.2 dWAR.

I bet that infield played hell with opponent batters' BAbips. Did anything get through?

Ignore dWAR--it is the devil

Ignore dWAR -- it's the worst movie Robert Forster ever made. And that's saying something.

dWAR now contains the position component of WAR, so the Career Leaderboard is now dominated by SS, C and other great defensive players.

Hey guys, let's take the handwaviest part of the formula, and make it dominate everything else!

Hey guys, let's take the handwaviest part of the formula, and make it dominate everything else!

I do find it curious that that we just assume that WAR components for positions are static and can not change.

dWAR, what is it good for?

I hate this. Keith Hernandez, Will Clark, JT Snow, Jeff Bagwell, Mark Grace, Don Mattingly, Wes Parker ... they all have negative dWARs. I realize that this all might make sense if you understand what the statistic really means, but come on. This statistic needs to be monkeyed with if anyone is going to trust it.

But if you want to know how good a guy was at his particular position there's Rfield isn't there?

Don Mattingly +33
Keith Hernandez +117
JT Snow actually comes out as -20.

EDIT: I've always ignored everything to the right of WAR on the B-Ref value ledger anyway (not that I mean it's not good information, but for my purposes I like the stuff to the left. The components that outline how you get to the final WAR. That's why I've never understood the complaints about what dWAR doesn't do a good job of measuring. It seems like the thing everyone is complaining about dWAR not providing is right there if they'd just pan over to the Rfield column)

The quick answer is that these numbers are always an interpretive estimate of what players contribute to winning. They're there to be criticized, not for "solace." And if somebody comes up with a better estimate, why shouldn't they abandon the old one?

The only people who should be put about by an upgrade in WAR formulas, as several posters above have noted with telegraphic snark, are those who had become addicted to looking at the single number in the WAR column and ending thought at that moment.

This. No matter who cranks it out, WAR is always going to be an estimate -- nothing more, nothing less. If the fundamental processes that go into the formula are valid (and I think that's true here once the RPW issues and perhaps SP leverage get worked out), then the results will be good estimates.

The specific rankings of certain players is really quite irrelevant. WAR is only supposed to throw players into large bins based on estimated value. From a team perspective, you don't sweat decimal differences between players -- you only care generally what kind of production you can get out of a guy vs. how much you're paying for him. If you think that any version of WAR is more precise or more useful than that, you're misguided about the purpose and function of the stat.

Just a heads up that with the help of Tango and Patriot I've reworked the runs to wins calculations, so the pitcher WAR is not as large as it was before. We were on the right track, but given we were using an approximation, we didn't realize how badly it would break down at edge cases like the Cy Young award winner. Anyways, the top pitching performance are now a little better than before, but not way better.

Post announcing it
http://www.sports-reference.com/blog/?p=675

If Player A has 0.5 WAR, and Player B has 1.4 WAR, then that is information I want to have. I don't want to see two 1's.

Assuming that one significant digit is the proper level of precision, it's not information. It's measurement noise masquerading as information.

With the WAR update, 23 of the last 50 NL MVPs go to pitchers, as well as AL MVPs. However, only 10 of the top 30 players in career WAR are pitchers.

This seems reasonable. The higher instability of year-to-year performance among pitchers relative to hitters would explain the both single-year and career outcomes. In a given year, the quasi-randomly distributed likelihood that a pitcher can have "everything go right" (health, hit luck/defensive efficiency, usage, control, etc.) and put up an MVP season makes the pool of potential MVP pitchers relatively large. Think of Mark Fidrych 1976, Mike Scott 1986, or Dontrelle Willis 2005. However, there is less outcome variance among position players. A defensive career year can help a guy luck into an MVP (Zobrist last year, Markakis 2008, Lonnie Smith 1989), but that's about it. Thus, the odds are significant that some Fidrych will have all the stars align for a single year, whereas the pool of sure-bet MVP candidates is relatively constrained to, say, the Pujols, Bonds, and A-Rods of the world. However, year-over-year, the position players are a better bet to come closer to repeating their previous performance, so only those pitchers who turn out to be similarly sure bets can do well in career WAR.

Thanks for looking at that James. I think that is reasonable as well. It may also be a residue of us using RA rather than FIP.