The WAR fallacy

[CA-ORIOLE]

I have a question. Let's say we fielded a 25 man roster of ALL ZERO WAR players. What would our record be? 0-162?

I've never been on board with the WAR stat.

I'd guess 40-45 wins or so.

[Fan4Life]

44-46 wins as that is thought to be replacement level.

Is there a 25 man roster stat sheet that shows what a replacement level team looks like statistically? For example, is there a replacement level 1b statistical example?

[33rdst]

Your rep just pushed me to 5 birds!

Great! Congratulations.

[Pedro Cerrano]

Is there a 25 man roster stat sheet that shows what a replacement level team looks like statistically? For example, is there a replacement level 1b statistical example?

[CA-ORIOLE]

Is there a 25 man roster stat sheet that shows what a replacement level team looks like statistically? For example, is there a replacement level 1b statistical example?

Go to fangraphs and sort out players by position and pick guys close to 0 WAR.

Example, VLAD (DH) Reynolds (3b) were full time players that were close to zero WAR last year.

[SilentJames]

Really. So past performance/statistics has no relevance to future performance?

One: Replacement level is not a steady thing from year to year, it shifts.

Two: WAR can have very large swings, see Zobrist, Ben WAR over the last four seasons: 1.4, 8.9, 3.9, 6.6

Three: Let me clarify, WAR is primarily and evaluator tool and people are using it incorrectly and putting WAY too much stock into it.

Its like this, you look at a players WAR over three or so years and you can usually get a pretty good idea of what type of player he will be, unless you look at a guy like Zobrist. But when you take WAR and just add it all together and say that this team will win X-more wins is a silly use of it.

Its like using a screwdriver to pound a nail in, something many of us have done from time to time but we should all know that there are far more appropriate tools.

Moreover it is just bad math. Eventually teams top out and wimply adding more WAR will have diminishing returns. You realistically can't make a 182 win team, we have only had one team in the last 85 years crack 110 wins. So just adding WAR means very little when you are already a top-team - diminishing returns.

[Pedro Cerrano]

One: Replacement level is not a steady thing from year to year, it shifts.

Two: WAR can have very large swings, see Zobrist, Ben WAR over the last four seasons: 1.4, 8.9, 3.9, 6.6

Three: Let me clarify, WAR is primarily and evaluator tool and people are using it incorrectly and putting WAY too much stock into it.

Its like this, you look at a players WAR over three or so years and you can usually get a pretty good idea of what type of player he will be, unless you look at a guy like Zobrist. But when you take WAR and just add it all together and say that this team will win X-more wins is a silly use of it.

Its like using a screwdriver to pound a nail in, something many of us have done from time to time but we should all know that there are far more appropriate tools.

Moreover it is just bad math. Eventually teams top out and wimply adding more WAR will have diminishing returns. You realistically can't make a 182 win team, we have only had one team in the last 85 years crack 110 wins. So just adding WAR means very little when you are already a top-team - diminishing returns.

2001 Mariners?

[Fan4Life]

Go to fangraphs and sort out players by position and pick guys close to 0 WAR.

Example, VLAD (DH) Reynolds (3b) were full time players that were close to zero WAR last year.

I'm not sure this works. Those examples are the result of what a replacement level player looks like based on the formula. I want to see the formula and determine how it arrives at a specific position's statistical example.

[DrungoHazewood]

I feel like I am reading too many discussions that go like this: "Last year we won 69 games, and if we add a player with X WAR, next year's team will only win about 69 + X, plus or minus a couple of games."

To me, this is an oversimplified and incorrect use of WAR. In the first place, the existing talent on the team is not static. Not everyone is going to produce at the same rate as the year before. Baseball history is full of examples of teams that won 20+ games more or fewer than the year before, with pretty small changes in personnel.

Whenever I post something using WAR there's an implicit level of uncertainty in that statement. When it's a projection that's a higher level of uncertainty. I think most people understand this. When you say Prince Fielder is a 4-win player, you're actually saying that he's been worth something like 4 wins per 600 PAs over the last X years, and that he'll probably be worth something like 3-5 wins a year over the next few years. It's just annoying to type that 23 times a day, and even more annoying to read it.

Second thing is that WAR is a useful way to measure a player's worth, but in my opinion it is wrong to think that the addition or subtraction of a player is a math exercise when it comes to computing effect on the W/L. A lot of times, a team just has certain needs, and if they are met, the effect can be much greater than a player's WAR value.

I think there's some small truth in there somewhere, but I'd be very careful of falling into the opposite fallacy that you can add just the right kinds of players and the synergies will combine almost magically to produce non-linear gains in wins. Gains, losses, slumps, streaks, luck, coaching, scouting... it's often thrown into a big bucket and narratives are generated that fit the outcome.

Now, I am not raising this point in order to argue that the Orioles will be contenders if they acquire Prince Fielder and Edwin Jackson, for example. I'm simply saying that baseball is a complex, interactive, non-static game, and not a simple mathematical exercise. (Thank God.)

Sure. But the best place to start is still with evidence, and your best evidence is often based on math.

[SilentJames]

2001 Mariners?

Actually it was hreet, I missed the 1998 yankees on that list and the 54 Indians. Point remains it doesn't happen often.

[weams]

Not at all its wins above replacement player...A replacement player is any player that SG says is freely obtainable....lol

A 0 WAR team would normally have 40 something wins.

Is it 48 wins at replacement? I thought it was.

[SilentJames]

I'm not sure this works. Those examples are the result of what a replacement level player looks like based on the formula. I want to see the formula and determine how it arrives at a specific position's statistical example.

And every year replacement level changes a little bit.

[SrMeowMeow]

One: Replacement level is not a steady thing from year to year, it shifts.

Two: WAR can have very large swings, see Zobrist, Ben WAR over the last four seasons: 1.4, 8.9, 3.9, 6.6 This is not unique to WAR. And it is caused by the fact that WAR takes as inputs a number of "simpler" stats, which are also subject to wild swings.

Three: Let me clarify, WAR is primarily and evaluator tool and people are using it incorrectly and putting WAY too much stock into it. I don't know what this means. WAR is a stat. You can project with it like any other stat. There's nothing wrong with saying "I project this guy for 2.5 WAR next season."

Its like this, you look at a players WAR over three or so years and you can usually get a pretty good idea of what type of player he will be, unless you look at a guy like Zobrist. But when you take WAR and just add it all together and say that this team will win X-more wins is a silly use of it. You can't rely on a team's 2011 W/L record as their 2012 baseline. But what does that have to do with WAR? The whole idea of WAR, which is borne out by experiment, is that you can add it up and approximate a team's wins for that season.

Its like using a screwdriver to pound a nail in, something many of us have done from time to time but we should all know that there are far more appropriate tools.

Moreover it is just bad math. Eventually teams top out and wimply adding more WAR will have diminishing returns. You realistically can't make a 182 win team, we have only had one team in the last 85 years crack 110 wins. So just adding WAR means very little when you are already a top-team - diminishing returns. Bad math? Is there any evidence of diminishing returns in WAR? I doubt it, because that makes no sense.

I disagree with a lot of this - replies in bold.

[DrungoHazewood]

I have a question. Let's say we fielded a 25 man roster of ALL ZERO WAR players. What would our record be? 0-162?

I've never been on board with the WAR stat.

WAR is designed to place a baseline under team performance. It says "if I don't try, and I just collect freely available talent to fill my entire roster, how many wins will I have?" Logically, that has to be somewhere around the level of the worst major league teams of all time, or at least fairly recent vintage. So that level has to be somewhere between 40 wins (the '62 Mets) and maybe 50 wins. So let's split the difference and call it 45.

Your zero value point then is the value equivalent of a whole team of players who could be picked up on the waiver wire at the end of spring training. Anything above that has a value in wins and in dollars.

[Fan4Life]

And every year replacement level changes a little bit.

Yes I just read one description of the RLP. While I think this idea works in a vacuum, I have a lot of doubt that it works well with regard to human beings. It relies on facts that are not in evidence. That the 31'st SS in a league of 30 teams with no SS's on the bench is in fact the 31st best SS and that the production created by that SS is the basis from which all other production for SS will be measured.

If we're talking about computers sure, measuring their ability has specific items to evaluate that do not fluctuate. Players are hardly this way. And, it also presumes that those evaluating who the 30 best SS's are, are infallible.

For this to work for me, you'd have to calculate what a team that achieves 48 wins looks like by position. This would also pigeon hole players to a specific production level for the position they play, but this is already a generally accepted philosophy.

Once I know what an RPL 1b looks like on a 48 win team, then I can know what my 1b is.