WAR, What is it good for?

[Filmstudy]

One thing I don't like about WAR is that they have re-baselined it to a .294 winning percentage. So WAR tells you how many more wins a team that was otherwise 48-114 (approximately) would have if they swapped in the individual's playing time. I am surprised by the number of people who think war is an increment from "average". That is part of the transparency problem with the metric. It's good for sophisticated baseball fans (like those on this site) to get a general impression of a mediocre player's ability relative to an estimate of the risk of being without.

Why .294 as opposed to .320 (a previous standard) or .333 (yet an earlier standard)? I think the .294 was determined so that there would be exactly 1,000 total WAR to be assigned in the league (given 162 games, 30 teams, it works out to 1001).

It might be reasonable to measure players by:

1. Wins relative to the league average at the position (WAA)

2. Wins above a floating replacement level that could be calculated each season at year end based on a standard of players who were not opening day starters (say)

3. Wins above replacement based on last season's calculated replacement level, calculated per #2

4. Wins above an arbitrary standard, in this case a .294 winning percentage

I think WAA makes more sense when comparing great players/teams, because the relative impact of great players stands out in bold relief with WAA and the effect is muted by WAR.

I also like WPA better, because it is much cleaner to me conceptually (and easier to explain) than the many calculations necessary to calculate WAR. The other good thing about WPA is that it eliminates most of the questions about the conditions under which something occurred. "Does WPA take into account the time I saw Chris Davis hit a 3-run HR with 2 outs in the bottom of the 9th to give the Orioles a 6-5 win?"...Yes, it does so by calculation in the difference in winning percentage before and after the event, so all clutch performances and failures are reflected and the leverage of each situation is explicitly included in the calculation.

[Filmstudy]

I like Tom Tango's characterization of WPA as a storytelling stat. Someone with an extreme WPA has to have some great (or awful) stories behind the numbers. But the leap from stories to skills is not really supported by much. Chris Davis is a good hitter, and good hitters often come through in big situations. Chris Davis doesn't necessarily have super powers in the clutch.

I like this too, but I don't see a lot of use of WAR on this site as a projection metric. Much of it is backwards looking, like the discussion of Jeter vs. Grich. If you're trying to answer the question "Who was better", then I think WPA and the explicit valuation of clutch performance is appropriate for hitters at a minimum. You could argue it unfairly overvalues outstanding relievers who get most or all of their opportunities in high-leverage situations.

[weams]

I like this too, but I don't see a lot of use of WAR on this site as a projection metric. Much of it is backwards looking, like the discussion of Jeter vs. Grich. If you're trying to answer the question "Who was better", then I think WPA and the explicit valuation of clutch performance is appropriate for hitters at a minimum. You could argue it unfairly overvalues outstanding relievers who get most or all of their opportunities in high-leverage situations.

Well, Jeter was almost a win better. But he needed to play those extra years to eke that out.

[DrungoHazewood]

I like this too, but I don't see a lot of use of WAR on this site as a projection metric. Much of it is backwards looking, like the discussion of Jeter vs. Grich. If you're trying to answer the question "Who was better", then I think WPA and the explicit valuation of clutch performance is appropriate for hitters at a minimum. You could argue it unfairly overvalues outstanding relievers who get most or all of their opportunities in high-leverage situations.

I don't agree that WPA says who was better. It says who performed better in higher leverage situations. But it leaves unanswered if that was a result of skill or luck. You'd expect variation in clutch performance and resulting WPA from random variation, and that variation would look an awful lot like what actually happens.

[VeveJones007]

I don't agree that WPA says who was better. It says who performed better in higher leverage situations. But it leaves unanswered if that was a result of skill or luck. You'd expect variation in clutch performance and resulting WPA from random variation, and that variation would look an awful lot like what actually happens.

Agreed. And you run in SSS issues by weighting on leverage. And some players will "luckily" get ABs in high leverage situations that they had 0 control over. An AB is an AB and an outcome is an outcome.

[InTheWinColumn!]

I like WAR when it goes our way...

[Filmstudy]

I don't agree that WPA says who was better. It says who performed better in higher leverage situations. But it leaves unanswered if that was a result of skill or luck. You'd expect variation in clutch performance and resulting WPA from random variation, and that variation would look an awful lot like what actually happens.

I don't buy the distinction. If you want to use it for projection, then I can see how you care why abnormal performance in high leverage situations skewed the result significantly, but if you are simply trying to answer the question "Who was better", I don't see any better method than WPA's reflection the actual results.

Here is my analogy from gambling. You go to Las Vegas to count cards for a week and come back $5,000 to the good. By your calculations for 32 hours played and considering your modeling of the rules/conditions, you should only have won about $2,000. Obviously, you got a little lucky. That might have been a matter of more attractive positive counts to bet, or a higher percentage of wins on your largest bets, or some combination. For looking backwards, the only number that matters is the $5,000. For looking forward, it's reasonable to delve deeper into whether that result was due entirely to luck, some conditions you weren't modeling, or some additional element that was exploitable. Going the other way, it's nice to have a model that says you should have won $2,000, but if you lose $3,000 that is the only number that matters looking backwards.

And yes, I'd expect variation in any set of results, but we're perfectly happy to revel in the long history of baseball statistics without much concern over the level of clutch performance (or full-season variation) that underlies them.

[Phantom]

I like this too, but I don't see a lot of use of WAR on this site as a projection metric. Much of it is backwards looking, like the discussion of Jeter vs. Grich. If you're trying to answer the question "Who was better", then I think WPA and the explicit valuation of clutch performance is appropriate for hitters at a minimum. You could argue it unfairly overvalues outstanding relievers who get most or all of their opportunities in high-leverage situations.

The problem with judging someone as better due to results as opposed to expected results is that then you start giving them credit for situations they didn't create. If you hit a three run homer in the bottom of the 9th inning down 2 runs, are you better than a player who hit a solo home run?

[DrungoHazewood]

I don't buy the distinction. If you want to use it for projection, then I can see how you care why abnormal performance in high leverage situations skewed the result significantly, but if you are simply trying to answer the question "Who was better", I don't see any better method than WPA's reflection the actual results..

The distinction is that it wasn't because of skill that you're giving credit. You're giving, for example, Derek Jeter credit for having his .840-whatever OPS in all those high-leverage situations his teammates created while not giving Aubrey Huff credit for his .840-something OPS because Daniel Cabrera put the Orioles in a six-run hole.

[DrungoHazewood]

The problem with judging someone as better due to results as opposed to expected results is that then you start giving them credit for situations they didn't create. If you hit a three run homer in the bottom of the 9th inning down 2 runs, are you better than a player who hit a solo home run?

Or, are you a better player for hitting a grand slam in the 6th inning of a 1-0 game, as opposed to hitting that same grand slam in the sixth inning of a 15-2 blowout? WPA says the first player is dramatically better.

Manny's grand slam the other day barely counted as a thing in WPA since the O's odds of winning before that were probably well over 90%.

[weams]

Or, are you a better player for hitting a grand slam in the 6th inning of a 1-0 game, as opposed to hitting that same grand slam in the sixth inning of a 15-2 blowout? WPA says the first player is dramatically better.

Manny's grand slam the other day barely counted as a thing in WPA since the O's odds of winning before that were probably well over 90%.

Any stat that weights those types of things is not really painting a true picture.

[Phantom]

Any stat that weights those types of things is not really painting a true picture.

I don't think truth has anything to do with it. Is RBI an untrue stat? Pitching Wins? Saves? Outfield assists? It's tough to find stats that don't factor in situations. That doesn't make them untrue so much as unhelpful and often useless.

[weams]

I don't think truth has anything to do with it. Is RBI an untrue stat? Pitching Wins? Saves? Outfield assists? It's tough to find stats that don't factor in situations. That doesn't make them untrue so much as unhelpful and often useless.

I can see if you are comparing it to a counting stat how you could say that. If the stat is used for comparison and it diminishes the impact of any grand slam, I'd have to say it becomes useless as you have stated.

[weams]

I don't think truth has anything to do with it. Is RBI an untrue stat? Pitching Wins? Saves? Outfield assists? It's tough to find stats that don't factor in situations. That doesn't make them untrue so much as unhelpful and often useless.

I never thought the stat itself was dishonest, just to use of it as an overall comparison metric. Thank you for the perspective though. Your case is well stated.

[DrungoHazewood]

Any stat that weights those types of things is not really painting a true picture.

Depends on what picture you want to paint.

I think that WPA's viewpoint has to be put in the right context. It basically says you can stop trying once the score differential reaches a certain point.