That's the asymmetry I was referring to.

Both metrics tell you how many more/fewer runs the player produced for his team than would an average player in the same circumstances. It's hard to imagine more symmetry than that.

The WAR model presumes that replacement level talent that will perform at the replacement level is available in the marketplace. It isn't

.

On average the group of waiver pickups and minor league free agents will play to replacement level. If they consistenly don't, then we just need to change replacement level.

Results will vary for every team dipping into this replacement player pool. Some wind up with players who can't field and bat .180 with no power. Others wind up with the 2012 versions of Brandon Moss, Justin Maxwell, or Justin Ruggiano.

Both metrics tell you how many more/fewer runs the player produced for his team than would an average player in the same circumstances. It's hard to imagine more symmetry than that.

No, they do not. In its simplest terms, fielding adjusts for degree of chance difficulty, batting does not. Rbat assumes all hitters faced equally difficult pitches, and produced hits and outs on equally difficult pitches.

Changing the scale of Rbat to runs above average to try to scale it to defensive runs saved is just a band-aid that shouldn't really fool anyone.

Fielding adjusts for degree of difficulty because the degree of difficulty varies widely. The degree of difficulty for hitting is much more stable.

On average the group of waiver pickups and minor league free agents will play to replacement level. If they consistenly don't, then we just need to change replacement level.

Assuming that's true -- and there's no reason to believe it isn't (*) -- it's still somewhat beside the point. Baseball franchises don't have infinite time (or infinite money), and there are frictions in player transactions. If you pick the wrong freely-available guy and he has a crappy 40 games and produces minus 1 WAR, the next guy has to produce .33 WAR per 40 games over the next 120 games just to get you back to 0 WAR for the year.

Results will vary for every team dipping into this replacement player pool. Some wind up with players who can't field and bat .180 with no power. Others wind up with the 2012 versions of Brandon Moss, Justin Maxwell, or Justin Ruggiano.


Yes, exactly.

(*) I'm assuming you mean, "The average expected production of freely-available talent over the next 162 games is 0 WAR."

No, they do not. In its simplest terms, fielding adjusts for degree of chance difficulty, batting does not. Rbat assumes all hitters faced equally difficult pitches, and produced hits and outs on equally difficult pitches.

This has nothing at all to do with your claim that the two metrics are in some sense on different scales, one measuring marginal runs while the other is measuring absolute runs. You are just completely wrong about that. Is it so hard to just acknowledge that?

The problem with measures like WAR as opposed to Win Shares is that WAR is trying to eliminate the luck, or part of the luck, from baseball. Luck is part of the game. Luck is a part of life, and a part of other sports and games like poker.

The WAR fans want to take a team that won 90 actual games, but 93 by adding up their component stats, and say that the result is better than a team than won 94 actual games, but 91 by adding up their component stats. I think that luck or no luck, the 94 actual win team achieved more of their goal than the 91 win team and should be rewarded accordingly by any rating system. Luck or no luck. There is luck involved in the component stats too. No one is precisely a 3.00 ERA pitcher or .300 hitter or 30 home run hitter. The actual ability is only somewhat around that number, subject to luck.

In real life it's impossible to play 1,000 simulations of a season. By the time one six month season has been played, players have gotten hurt or old and younger players have matured or added ability. There is no steady state. That Team A beat Team B in run components doesn't help the team for next year, since no value can be banked and the players on the two teams won't be in the same condition anyway.

Assuming that's true -- and there's no reason to believe it isn't (*) -- it's still somewhat beside the point. Baseball franchises don't have infinite time (or infinite money), and there are frictions in player transactions. If you pick the wrong freely-available guy and he has a crappy 40 games and produces minus 1 WAR, the next guy has to produce .33 WAR per 40 games over the next 120 games just to get you back to 0 WAR for the year.

What point are you talking about? All player performance is variable. Just because a guys has been an above average player in the past doesn't guarantee that he'll play better than Jason Bay 2010-12 once you sign him. Or everything breaks right for a player who is usually average and he puts up an MVP level season.

That is totally separate from the concept of replacement level.

At its most fundamental level, Miguel Cabrera stood where a "third baseman" usually stands, with minor adjustments by his manager, and turned balls that would have been hits had he not been standing there, into outs. He, therefore, prevented runs.(*) How could his fielding value, on the same scale as his batting value, therefore be negative? That fundamentally doesn't work.

None of this is true. There are three components going on here. One is offensive runs relative to the league average player, which is simple. Another is defensive runs relative to the average player at your position in that year (UZR, TZ, whatever). The third is the value of his position relative to the value of the average player's position. Because 3B is harder to play than LF or 1B, Cabrera gets a slight bump (about +3 runs over a full season). Because his defensive relative to position was poor (as measured by UZR, TZ and really all of the scouting reports and common sense), he was a slightly *below average* defender, which of course is not the same thing at all as saying he was worse than a nonexistent defender at his position.


KT's point is the most salient and unfortunately CFB did lawyer it up. KT is not saying that WAR measures wins perfectly, he's saying that WAR is trying to describe wins - that's unit it's going for, obviously. He's also right that every transaction in baseball history has been made because team A thinks they will win more games either now or in the future because of the transaction. Obviously there are other components to wins than just WAR, but estimating wins is the necessary language of running a baseball team. WAR, as it's currently constructed, does a good job of measuring the context neutral, statistical value a player offers. What's left is to estimate the value of character, clutch playing, etc, but that still needs to be done on the same scale (runs to wins).

Think about it, if you trade Milton Bradley for peanuts aren't you saying that his character is worth -2 wins (or something)?

Risk-free replacement level performance is not available in the marketplace

Neither is benefit-free replacement level performance.

Random variation -- catch it!

And folks, please stop falling for SBB's main sucker punch. All baseball projection has "high" variance because baseball (like any 0/1 event) has a large amount of RANDOM variance.

Go ahead, project how many heads you will get in 162 flips of a coin, flip it 162 times and see how close you come. We know that, on average and assuming a "fair coin", you'll get 81 heads. But the probability of getting exactly 81 heads? That's only 6%. There's a 10% chance you'll get 90+ and therefore also a 10% chance you'll lose 90+.

So even if your fave projection system right now could completely accurately predict the true talent level of the team right now, all it can possibly achieve is to predict the non-random variance and it is always going to have to deal with a random standard deviation of about 6.5 wins.

Similarly, if God him/her/itself tells you a player has a true OBP of 350 and will get 600 PA, you still have a random standard deviation of 2%. Now 2% is actually pretty small but of course 2% is 20 "points" of OBP and there's a "big" difference between a 330 OBP and a 370 OBP. That's nearly 2 wins right there. There's nothing that can be done to explain that variation.

Let's be very clear -- that random variation is not due to the quality of pitcher faced or whether a guy played more often with the wind blowing out or even whether he amped up before the game. Those are things models could (theoretically) control for but God can tell you before each PA what the guy's chances are in that PA and it's still a coin flip (with whatever p(success) God told you).

So you have a simple choice. Throw up your hands and say the whole damn thing is so variable you might as well not analyze it at all. Or do your best to explain the parts that can be explained while understanding that there will always be large random variation out there.

Or buy whole hog into the notion of predestination but then Snapper is gonna get on your ass.

Or buy whole hog into the notion of predestination but then Snapper is gonna get on your ass.

But only because he was predestined to get on your ass.

It's interesting to think of a team composed truly of replacement players. I mean, the 2012 Astros come damn close.

I pulled up a list of teams with the most players with 1 WAR or less and 250+ PA. The Oakland A's in 1979 were the only team with 12 players:


Those A's had a record of 54-108. Their regulars:

WAR POS Player . . . Age PA
1.5 1B Dave Revering* 26 513
0.9 C Jeff Newman 30 552
0.7 CF Dwayne Murphy* 24 489
0.6 C Jim Essian 28 352
0.5 3B Wayne Gross* 27 521
0.2 RF Tony Armas 25 297
-0.6 OF Mike Heath 24 286
-0.9 DH Mitchell Page* 27 539
-1.0 LF Rickey Henderson 20 398
-1.0 SS Rob Picciolo 26 363
-1.4 RF Larry Murray# 26 261
-1.6 INF Dave Chalk 28 250
-1.7 2B Mike Edwards 26 422

Let's be very clear -- that random variation is not due to the quality of pitcher faced or whether a guy played more often with the wind blowing out or even whether he amped up before the game. Those are things models could (theoretically) control for but God can tell you before each PA what the guy's chances are in that PA and it's still a coin flip (with whatever p(success) God told you).

Hal Richman's Strat cards can tell you before each PA what the chances of every result are and Wes Parker can still OPS 150 points below real life in an as-played, auto-transactions 1969 replay.

In real life it's impossible to play 1,000 simulations of a season.

The funny thing is, reality isn't a simulation, and far too many people seem to be forgetting that fact these days.

The funny thing is, reality isn't a simulation, and far too many people seem to be forgetting that fact these days.

It's a real shame too. I haven't really had a troubling nightmare in ages due to the beauty of the quick save. Anytime it looks grim I can just F5 it and know that if I don't make it out of there I can go back and try out the next permutation.

I suppose if you start and stop enough jobs/relationships you could get away with treating life like a series of simulations.

The WAR fans want to take a team that won 90 actual games, but 93 by adding up their component stats, and say that the result is better than a team than won 94 actual games, but 91 by adding up their component stats. I think that luck or no luck, the 94 actual win team achieved more of their goal than the 91 win team and should be rewarded accordingly by any rating system

But nobody's saying that at all (and if they are, they should stop). As Walt and several others make clear, we have no idea whether a "93-win" team is "better" than a "91-win team" by any measure, including actual wins. That's just a terribly small margin. As is 94 > 90, for that matter. And the only rating system that reality cares about is actual wins, anyway, so there's hardly a problem with that. The division title will go to the 94-actual-win team without exception.

I totally agree that watching actual games to see who wins is the whole fun of sports, but it's still very interesting to see how much luck might have been involved in getting there. It helps us think about that 1969-Mets experience we've all had where we wonder "how are those guys doing it?"

Go ahead, project how many heads you will get in 162 flips of a coin, flip it 162 times and see how close you come. We know that, on average and assuming a "fair coin", you'll get 81 heads. But the probability of getting exactly 81 heads? That's only 6%. There's a 10% chance you'll get 90+ and therefore also a 10% chance you'll lose 90+.

Real life is less variable because the chance of winning every game is not 50%. If you are simulating results of a hitter, hitters as a group have different results depending on what the count is. So if you do a pitch by pitch simulation, there will be less overall variance because of the situational difference.

If instead of flipping one coin with a 50% chance, you flipped two coins, one with a 75% chance of heads and one with a 25% chance of heads, the outcome will still be centered at 50%, but there will be less variance.

Wow, just a personal aside. I remember spending part of the summer of 1979 in the Bay Area and going to several Oakland A's games. You literally could drive up to the park just before game time (easy parking, no lines, no crowds, etc.). Many games only had around 2,000 in attendance. Boy, those A's were bad.

It is actally unknowable whether or not "real life" that we perceive is, in fact a simulation. In fact, it could be simulations all the way down.

But this is a dodge anyway. All models are wrong, some are useful.

Dammit RNJ, I was counting on you to get this one:

Eventually someone is going to come up with the proper rbi stat, which is to assign .25 bases to every remaining base of the runners on, including a full 1 for the batter, and then track batters advanced, batters scored, and total opportunities. Until that day arrives you are going to have the stat nerds doing everything with the expected run scoring matrix, and not one normal person giving a rats ass about it.

James Tuttle did this 15+ years ago in rsbb

Sam Miller's WAR article in ESPN the Magazine was horrendous, which is typical for the magazine. However, the articles on Michael Jordan and Bob Voulgaris are worth reading.

Sam Miller talks about WAR being so scientific, but requiring faith. What the hell is he talking about? Some might think this article is another about sabermetrics being more advanced than traditional scouting methods. It seems to me that this is more about ridiculing those who don't believe in evolution, vaccines, or global warming (or is it climate change?)

Good science should stand up to questions and scrutiny. Being more complicated and harder to understand doesn't demand faith. It demands more research and more work.

WAR is useless. It is an extraneous number. Mike Trout's 10.7 just adds confusion. Yes, his number is bigger, but it doesn't have meaning. I mean it's an abstract number. Couldn't you tell who was better with basic stats and common sense? Their OBP SLG AVG lines were very similar, but Trout is a great CF with blazing speed on the bases. Cabrera is a poor defensive player who runs like he has a piano on his back.

More importantly, WAR is derived from the basic stats. OK, there is a park factor and other defensive ratings thrown in. However, the park factor is not reliable. Different types of players are affected differently by the park. Lefties, righties, power hitters, speedsters, ground ball pitchers, pull hitters, etc. all vary greatly in how they are affected. Then you have 5-year regressions. Some parks are altered in that time. If a player joins a team that has a favorable PF, but the most recent year is a down scoring year, then he is penalized because the previous years favored scoring. Some of the defensive metrics are entirely subjective. Hard to tell where zones end and begin. A line drive between 1st and 2nd. Whose zone, and does it count since it was a rocket shot?

And you know what? I'd like to see sabermetrics used in the vaccine debate. Analyze the data. Did you know that, according to the CDC, all polio cases in the USA in the last 30 years were caused by the vaccine?