THE BOOK--Playing The Percentages In Baseball

I don’t quite follow the notion of adding a value for “Replacement”, which you seem to be calling “Playing Time”.  The Fangraphs glossary is not much help… can you explain?  I don’t understand how we can add “Replacement” into an equation and come up with a number that is “above Replacement”.

The other components are runs above average.  So the part called replacement, or playing time, is to account for the difference between an average player (a second baseman with a 100 OPS and -2 glove) and a replacement level player (the guy playing 2B in AAA).

You can leave this part out if you want, just know that you are figuring wins above average instead of wins above replacement.

All the other stats are “runs above average”.  Imagine you have two players, one who is +1 runs in 300 PA, and another who is -1 runs in 600 PA.

The first player may have performed better, giving his playing time, but he wasn’t there for 300 other PA.  Someone ELSE was.  That other guy, that schlub, was worth about -10 runs in those 300 PA.

So, given that each player was offered 600 PA, and the first guy played for +1 in 300 PA and didn’t show up (or wasn’t given a chance) on the other 300 PA, we assign him -10 for those 300 PA.

His final score is -9 runs compared to the other guy of -1 runs, or a gap of 8 runs.

Alternatively, you can just add +20 runs per 600 PA to both players.  So, the guy who is +1 in 300 PA gets an extra 10 runs, and comes in at +11.  The other guy was -1 in 600 PA, so we add +20, and he comes in at +19.  That’s a gap of 8 runs.

Got it?

Yes, makes sense, thanks guys!

I feel like it would make more sense to do fielding and offense above replacement.  I think “above replacement” is a much easier concept to understand than “above average”, when looking holistically at players.  But, their site, their call!

Mike, no you would be wrong on two fronts:

1. Average is completely understandable.  Replacement is not.  It’s subject to whatever baseline level you choose.  Average is average.

2. There is NO SUCH THING as hitting runs above replacement and fielding runs above replacement.  This was a problem that plagued WARP from its inception, and only recently was disavowed.  We’re not going back down that road.

WAR’s framework, as its laid out, is perfectly usable, understandable, and adaptable.

You change anything in there (hitting above “replacement"), and you are going to have a mess on your hands in terms of adaptability, and possibly understandability.

It’s better to understand why we think WAR works, than for you to think it might not work, and proposing a different solution, especially one that we’ve already considered, studied, and rejected.

This is probably a dumb question, but I’m curious.  Supposing there was a major league team where every player on the roster was exactly at replacement, had a WAR of 0.  What would be their expected record?

Look at the Pirates for your answer (team WAR for 2010 very close to zero on BB-ref).

Their expected record is whatever the replacement level is set at.  Rally sets it at around .333 I think.  I set it at .300.  BPro sets it at .250.

I think you are justified at anything between .250 -.350.  You might be able to argue .200 or .400, but it’s a much tougher argument.

A team of all replacement level players would win around 48 games in theory.

According to Rally’s theory, it’s 54.  My theory says 48.6.  Clay’s theory says 40.5.

You can reasonably argue lots of numbers via whatever theory you want to provide.

Let’s not say my opinion is wrong, please.  I have a very clear understanding of a replacement player.  The Red Sox field 3 or 4 of them every night.  I have no idea who (if anyone) is an “average” player in the Red Sox starting lineup.

I don’t understand why there’s no such thing as hitting runs above replacement… but if you say it’s been hashed out, I guess the onus is on me to look into the history of it.

For what it’s worth, when I say “intuitive” or “understandable”, I am not talking about you or anyone else on this comment board.  I am more talking about the general population that enjoys watching baseball.

Once you set whatever replacement level you like, there is no wiggle room for how many wins a team of these replacement players would have, right? 

The wiggle room is whether you set replacement offense to -20 per player or -15 or -25 (per 162 or per 150 games) and whether you set replacement pitchers to one run or 1.5 runs (or whatever) (per 9 IP) worse than average.

Fpr example, if I set all my offensive players to -20 runs (per 162) and all my pitchers to 1.25 (per 9), then I am exactly defining how many wins my team has (around 43 wins I think)…

@Mike: You can see that replacement level is differently defined depending on the system. Tango himself mentions that three different types of WAR/WARP run different replacement levels. However, the average hitter is very clearly defined based on the league’s overall numbers. Therefore it’s easier to compare to an average baseline and then add on value above replacement of your preference rather than taking the components over replacement.

That’s one of the ways I think of it, at least.

Baseball-Reference has a good blog post today on this year’s Team WARs using their WAR system.

http://www.baseball-reference.com/blog/archives/8225

Mike, yes, it’s a long history.  But here are two main smmary points:

1. There are replacement players, not replacement hitters and replacement fielders.  Unless you’re talking about specialists with little playing time, a player has to both hit AND field (or take up the DH spot, forcing everyone else on the team to field).  A run is a run, so whether a player is better in the field or better as a hitter doesn’t really matter—it just matters how good he is overall.  Thus you measure his offensive prowess, his defensive prowess, and then report his value over replacement as a combination of the two.

2. Actual replacement-level players tend to be average fielders and bad hitters (although there’s a continuum, certainly).  From this perspective, you could call average defense replacement level defense and “average minus 20 runs” replacement level offense.  (This is the VORP approach, in a way, except they define replacement level hitting differently for each position and don’t use a separate position adjustment.)

"That’s one of the ways I think of it, at least.”

That’s THE one way I look at it.  Everyone has the same “average” standard: average.  Replacement is added on top.

Though I should say that “average” might be a bit different depending on how you handle pitchers-as-batters, as well as the AL-v-NL differences.

I like Tango’s .300 repl level the best. Replacement is a fuzzy concept, as opposed to average. So why not keep it as round, simple, and elegant as possibe.

Not only is .300 nice and round, but you can ‘derive’ it from other nice round values. Baseball is built around multiples of three--3 outs/inning, 9 innings/game, 9 lineup sots, etc. So, repl is -18 runs per full slot per 162 G, with 9 full slots on offense and 9 on defense. So, (9 + 9)* 18 is -324 runs for the team. Divide by 10 R/W to get -32.4 wins from average. Which is exactly a .300 win %.

In terms of “nice round numbers”, you can also say that replacement level would be one fewer run scored and one more run allowed.  So, 3.5 runs scored, and 5.5 runs allowed.  PythagenPat says that’s almost exactly .300.

Indeed, if you make the “x” part of pythagenPat as .286, it’s exactly .300.

9^.286 = 1.8746

(3.5/5.5) ^ 1.8746 = .42857

And .42857/1.42857 = .300

Another thing that I love about WAR is that you can put all players, pitchers, non-pitchers, starters, relievers, etc, and list them on one scale, like say on the 1994 Expos. You might get a surprise here or there, but overall, it conforms to expectations. And if it does that, it’s alot easier to trust for teams you are not too familiar with.

Expos:

http://www.tangotiger.net/wonloss/index3.php?teamid=MON&yearid=1994

I’m sure this has been explained somewhere, so feel free to point me to a relevant article rather than type out a separate explanation here, but: why was the concept of replacement level added to the practical evidence of the average? Was it just a way to relate performance to salary?

Personally, I usually just look at RAA and leave out the whole theoretical replacement level.

It’s a very simple question, e poc: Who is more valuable, a guy with 0 RAA in 30 PA, or a guy with 0 RAA in 300 PA?

Once you answer that (and it’s an obvious answer), you should figure out pretty quickly why RAA is not always the right way to do it.

"why was the concept of replacement level added to the practical evidence of the average? “

What Colin said.  At the career level, the earliest form of the player lists I put on Baseballprojection, their original inspiration, were Pete Palmer’s lists in the Hidden Game of Baseball.  At one point Glen Hubbard ranked higher on the list (might have been later, in Total Baseball) than Pete Rose.  Part of that was the use of a flawed fielding metric, but mostly it was the playing time.

Bill James had a chapter. “Rain Delay” in the 1988 Baseball Abstract where he pretty much laid out the format for WAR.

WAR and WAA are EXACTLY the same thing!  The only reason we have the concept of replacement level is to be able to attach a dollar value to a player.  That is the ONLY reason.  If we don’t care about that (dollar value) then there is no need for a “replacement level” metric.  That is not to say that it isn’t just fine (to use WAR or RAR) even if we are not interested in a dollar value.  As I said, WAR and WAA are exactly the same thing, just as long as if you are comparing two players you are using the same replacement level.  People who have a problem with a metric that uses replacement level as the baseline simply do not understand the concept.  Now, if two people are using a replacement level metric and they don’t tell you what replacement level is, compared to average, then you may not be comparing apples to apples.

And, by the way, you HAVE to use replacement level to determine dollar value.  There is NO dollar value that you can attach to a level of performance or talent unless you figure out the average performance/talent of a player who plays for the league minimum.  And that IS the definition of replacement level (even though we don’t know exactly what that is and it varies from position to position, from year to year, and from team to team)!

Replacement level predates that Rally.  It was either when James compared Clemens/MAttingly and Rice/Guidry (I guess that would be the 87 Abstract). 

It could also be even earlier, when James compared the win% of players and said “what’s the chance a .400 player would put up those W/L records”, where the implication is that it would be harder to put up a 20-10 record than a 10-5 record.

That would be the genesis for replacement level talk (for me).  After that, you had Woolner and VORP, and that spawned a TON of repl talk at the old Fanhome.

To me, it all made sense when I introduced the fielding spectrum adjustments. 

***

I agree with MGL that the major value to repl talk is about money (any by extension production).  There’s a minor value if we are talking about guys who get injured alot, or not at all.

The Big Bad Baseball Annual also picked up the replacement level mantle between James and Woolner, and was the first to use the “WAR” name that I am aware of.  It was based on Offensive and Defensive W% and IIRC it didn’t really have a position adjustment to speak of, but it was certainly a replacement-level metric.

Win Shares to would be repl-level based, sort of.

I never read the BBBA until I picked up a copy a few years ago.

I disagree with that WAA and WAR are “exactly the same thing” and that the only reason to use replacement is for dollar value.

More important is the playing time issue, as Colin and Rally have said.  IMO, one should never use a metric with average as its baseline because that implies that an average performance has no value. That creates big issues when dealing with players of different playing time.

Just because you see a number that says “0” doesn’t mean no value.  Other people may imply that, but the metric doesn’t.

As I said:
“about money (and by extension production)”

If you have someone who performed average in 600 PA or in 6000 PA, that is NOT the same thing, and MGL isn’t saying it is either (I don’t think). 

However, if a .700 performance given 250 IP is as indicative of talent as a .550 performance given 500 IP (numbers for illustration purposes only), then the talent level might imply say .535.

And so, that’s what you tack on to both player, that he’s a .535 player.  You have on guy who is a .535 player (+.035 wins per 9IP) who performed at .700 over 250 IP, and you have another who is .535 who performed .550 given 500 IP.

Having a .400 replacement level makes the performance of one guy at +.300 wins per game given 250 IP, and the other guy’s performance as +.150 wins per game given 500 IP.  Their overall WAR is identical.

That’s really what the value is of PERFORMANCE-based measures with replacement level, is that it’s, in a roundaboutway, a way to regress performance to link to true talent.  That is, what Bill James was doing in the 1983 or so Abstract.

But, once you have a talent-based metric, then playing time is irrelevant.  (Except of course if there is talent is getting, or not getting, injured.)

Use of replacement level is about money, and about production.  But, you don’t necessarily need replacement level as a concept to exist.

For something that is supposed to be objective, there are so many subjective components. Park factors, replacement level, positional adjustments and defensive metrics are all subjective variables, so how can the output be objective?

I also don’t get the point of cramming something different into the “framework”. A metric isn’t a framework; it is exact.

Fiddling with the numbers may make for an interesting personal journey, but you can’t expect everyone to follow along as you figure it out.

For something that is supposed to be objective

No, it’s not supposed to be objective.  It’s supposed to be reasonable and consistent.

A metric isn’t a framework; it is exact.

No, a metric is not exact.  A metric IS a framework, or, at the least, it’s a way to measure something.

Fiddling with the numbers may make for an interesting personal journey, but you can’t expect everyone to follow along as you figure it out.

No one is asking everyone to follow along.  The only thing being asked is to understand the thing before denouncing it.  Don’t denounce it because you either don’t understand or don’t want to understand.

A metric is a “standard” of measurement. It is clearly defined; not open to interpretation. When something is a foot, we all know exactly what that means. There aren’t interchangeable variables that go into determining the value. From a baseball perspective, we know what ERA is. It is earned runs divided by innings pitched. There is no debate about the components.

WAR is a fine concept, but if the components can be altered, then it is not one metric. Furthermore, if the underlying variables are not consistent and reasonable, then the output can not be as well.

Finally, while something not understood should not be denounced, it can certainly be dismissed. If someone wants to be taken seriously, they need to make the case for their argument. Proclaiming superiority isn’t very convincing, and if the purpose isn’t too convince, then the argument is really irrelevant.

Just curious - when calculating WAR, do you use a scale/system/formula to make sure that pitcher replacement level is equivalent to position player replacement level?  MGL in #13 had a couple of different formulae as examples of replacement level, and I wasn’t sure if the two were linked or arrived at independently.  If the latter, it seems to me that we should be careful when comparing WAR between pitchers and hitters.

WAR is a framework.  I don’t even know where the metric thing started or why you keep bringing it up.  Since you want to beat the horse:

http://en.wiktionary.org/wiki/metric

A measure for something; a means of deriving a quantitative measurement or approximation for otherwise qualitative phenomena (especially used in Software Engineering)

“Metric” is NOT some indisputable single meaning definition.  WAR is a metric under some definitions.  It is a FRAMEWORK as I describe it.  The various implementations of the WAR framework is a metric.

***

Finally, while something not understood should not be denounced, it can certainly be dismissed.

If you meant ignored, then yes, I agree.  If you mean dismissed, then I disagree, since dismiss means that you’ve considered it and come to a conclusion.  Let’s presume you mean ignored.

If someone wants to be taken seriously, they need to make the case for their argument. Proclaiming superiority isn’t very convincing, and if the purpose isn’t too convince, then the argument is really irrelevant.

Are you just making stuff up?  Who is “proclaiming” anything?  I’ve made my case for WAR throughout my blog.  There is no “proclamation” as if I’m giving my opinion without evidence.

Other people may proclaim it because they trust me.  You can take it up with them.

Just curious - when calculating WAR, do you use a scale/system/formula to make sure that pitcher replacement level is equivalent to position player replacement level?

I set it so that about 43% of the WAR goes to pitchers, and 57% to non-pitchers. 

This is based on my model of the standard deviation of talent among non-pitchers and pitchers.

It does NOT necessarily apply to all of MLB history.

Btw, Preston, great question.

With respect to subjective and objective, ERA is a horrible example, for reasons that I trust are obvious.

It is impossible to construct any sort of metric that does not include subjective choices by the creator.  But *once those choices are made*, the results that follow are objective, arising from the formulas that have been set forth.

The distinction is important, because it is the difference between someone designing a formula to produce a desired result or taking a logical approach, one path out of many possible ones, and following where it leads.

Right, the objective component of any particular WAR metric is that you can document your methodology, and it’s repeatable - it applies the same to everyone, and someone else can take it and assuming they use the same assumptions and methods they can get the same results.

The question is if its desirable for everyone to use the same assumptions and methods.

@36 I am not referring to your use or presentation. Like or not, sabermetric “frameworks” are presented in the mainstream with arrogance. That isn’t going to convince most people, especially when there is so much subjectivity invovled.

Also, while I did mean ignored, the concepts can also be dismissed on a theoretical basis. One might not understand how WAR is calculated, but they can certainly discredit a “framework” based on flawed variables (i.e., if one thinks the concept of replacement level is flawed, then WAR becomes flawed regardless of how it is calculated).

Yes, Will, there are people in this world who think if something isn’t perfect it can’t be good, and on the other side there are people who believe that is something is good that it has to be perfect.

The truth is, of course, somewhere in the middle.

Will, yes, if you can discredit replacement level, then you can dismiss WAR.

It’s a very simple question, e poc: Who is more valuable, a guy with 0 RAA in 30 PA, or a guy with 0 RAA in 300 PA?

Once you answer that (and it’s an obvious answer), you should figure out pretty quickly why RAA is not always the right way to do it.

Forgive me for being a bit behind the curve, but I don’t understand why this makes a difference. You could easily argue that both guys are equally valuable, in the same way that you might argue that a guy with .5 WPA in 30 PA is more valuable than a guy with -2 WPA in 300 PA. If you want to discover true talent rather than value, you need to properly regress, in which case (assuming the only thing we know is the two samples you’ve given) you still can’t tell which is the better player.

It’s the same in the case of WAR, I think (but I’m not totally clear on this, which is why I’m asking). If one guy has 2.0 WAR in 100 PA and another guy has 2.0 WAR in 600 PA, you could say they were equally valuable (with respect to replacement) or that the former was more valuable (with respect to average). If you want to figure out true talent rather than value, you have to properly regress, etc.

It seems like basically the same process in either way, which is why I’m not clear on why replacement would be a better baseline than average.

And, by the way, you HAVE to use replacement level to determine dollar value.  There is NO dollar value that you can attach to a level of performance or talent unless you figure out the average performance/talent of a player who plays for the league minimum.

This makes intuitive sense, but why is it necessarily so? Either way, the market is going to determine what FAs are paid each winter, so empirically it seems like you could equate salary to average, with a scale that went in both directions from that salary. For instance, in this coming offseason we might see that an average player signs for $10M a year and that each run above or below average is worth +/-500K. Instead of getting into the hypothetical discussion of replacement, teams would make decisions based on the actual replacements they have and their actual salaries and their projected performances above or below average as well as their projected durability.

Forgive me for being a bit behind the curve, but I don’t understand why this makes a difference. You could easily argue that both guys are equally valuable, in the same way that you might argue that a guy with .5 WPA in 30 PA is more valuable than a guy with -2 WPA in 300 PA. If you want to discover true talent rather than value, you need to properly regress, in which case (assuming the only thing we know is the two samples you’ve given) you still can’t tell which is the better player.

This has nothing to do with telling who the “better” player is. It has to do with the understanding that playing more, in and of itself, has inherent value. You can offset that value through especially poor performance, but that doesn’t mean there isn’t value in playing more.

#44, if the average player makes 10mm, how would you know how much to pay a player that is 1 win less than average without establishing a replacement level?  Replacement level performance establishes the level of performance for which you pay around the MLB min.  Average salary tells us how much you pay per win.  Now, it might not be linear (although we think it is based on the evidence), but regardless you still need to determine the performance of players who are “freely available” and will work for the MLB min, in order to establish any salary guidelines whatsoever.  Now, just because an average player makes 10 mil and he is 2 wins above a replacement player in talent does not necessarily mean that a 4 win player is worth 20 mil, but it is a good assumption.  The reason that linearity is a good assumption is because two 2-win players on a team is pretty much exactly the same as one 4 win player, assuming everyone else is equal of course.  Now in reality, the $value of a player is strictly supply and demand, like any other commodity, so it is possible that it is NOT exactly linear, but again, linearity is a pretty good model for talent versus salary, and, as it turns out, is probably correct.

The presumption being made regarding the win rate stat of a player (let’s call it win%) and his value is that playing time follows win%.

Let’s say you have these players:
.380 win%
.440
.500
.560
.620
.680

Playing time should follow similarly.  So, it would be like this:
.380 100 PA
.440 200 PA
.500 400 PA
.560 500 PA
.620 600 PA
.680 700 PA

or some such.  That’s why we only need the rate stat, since we presume that playing time will match it.  We don’t expect to see something like this:
.380 100 PA
.440 500 PA
.500 300 PA
.560 200 PA
.620 700 PA
.680 600 PA

In that case, we WOULD need to know about replacement level.

To the extent that injury is a skill separate from talent, then that’s why we would want to know about PA.  And, you can make a good case of that, if you want to say it’s easier to lose playing time if you are 37 than 27. 

The best thing to do is stick with WAR, because it’s easier and safer.  Just be a bit careful.

@e poc: As Colin has more or less intimated, playing more is inherently valuable because teams have to replace playing time missed at a position with someone. So the guy who was worth 0 RAA in 30 PA had to have someone replace him for the remaining 270 PA to match the guy with 0 RAA in 300 PA. That “replacement player,” provided a team paid no additional value past the league minimum, is probably significantly worse than 30-PA guy. 300-PA guy provided value just by playing.

Think of a more extreme example. Who was more valuable to the White Sox, Juan Pierre or me?

Pierre: 143 G, 655 PA, -1.1 RAA (offense and defense) according to FanGraphs
Me: 0 G, 0 PA, 0 RAA

Pierre was worse than average this season, while I was average by default (you can use someone with 3 or 4 PA for a more realistic view), but clearly Pierre had more value to the White Sox than I did.

Value = “how much would you pay the guy for that performance?”

I tend to get wrapped around the axel when I think about this too much.  Colin implied that by basing WAR on replacement as opposed to average, we are able to get around the problem of valuing a guy with 0 RAA in 30 PA vs. a guy with 0 RAA in 300 PA.  I see how it accomplishes that. 

So my questions is… how do you value a guy with 0 RAR in 30 PA vs. a guy with 0 RAR in 300 PA.  They both end up with the same WAR.  So does it follow that they have the same value?

"They both end up with the same WAR.

You mean the same WAA.

No, they don’t have the same value.  Value is based on either expected playing time, or actual playing time.

Based on actual playing time, the guy with 300 PA has more, so he’s worth more.

Based on expected playing time, the guy with 0 RAR in 30 PA is more likely to be a replacement level player than the guy with 0 RAR in 300 PA.  So, the guy with 30 PA has a lower true mean and therefore a lower expected playing time.

You mean the same WAA

No, I actually meant WAR.  I swithced from RAA (runs above average) in Colin’s original example to RAR (runs above replacement).  I expect that the identical RAR implies identical WAR.

I follow what you’re saying about the 30 PA guy being more likely to be truly replacement and thus got less playing time, but that’s speculative.  The performance we’re considering is past history and we’re trying to value that, no?  Perhaps a more intriguing question is how does one “value” a guy with 1 WAR and 200 PA vs. a guy with 1 WAR and 300 PA?

Sorry, I misread you.

“So my questions is… how do you value a guy with 0 RAR in 30 PA vs. a guy with 0 RAR in 300 PA.  They both end up with the same WAR.  So does it follow that they have the same value? “

Yes, same value, regardless if 1 PA or 10,000 PA.  Those players should get non-tendered every year.

Tango, that’s for replying to my playing time comment, but it was addressed to MGL’s post, not yours.  And he clearly said that dollar value is the only reason to consider replacement level.

Your statement:

Just because you see a number that says “0” doesn’t mean no value.  Other people may imply that, but the metric doesn’t.

I understand what you’re implying--that a scale is a scale regardless of the zero point--but that’s not how such a metric gets used or interpreted.  People add these figures up.  They directly compare players with six at bats vs. those with 500 at bats. In fact, I am not really interested in a metric in which I can’t do those things.

Replacement level is critical for proper evaluations across playing time levels, not just for dollar valuations.  IMO, it needs to exist.

If, as Colin/45 suggests, this has nothing to do with true talent, and is entirely about value, then we are already implicitly talking about money. And if, as I initially asked, the whole reason for replacement-level theory is so that we can judge salaries, that makes sense to me. I don’t have an answer to MGL/46, so I guess it makes sense that you need some sort of sub-average baseline for the purpose of financially valuing players. The free market is itself entirely theoretical, so it makes sense that to predict and act within that market theoretical frameworks would be necessary.

If you did, however, want to talk about which players were better than others, I think, as MGL suggests in 26, that RAA is just as good as RAR, and I’d suggest it’s even better, since it’s entirely empirical. For that reason, I think it’s also probably more likely to catch on with mainstream fans/media.

MGL, you said the only reason to use WAR instead of WAA is to attach a dollar value to the player, but it seems to me you need to use WAR instead of WAA to compare how much value two players provided over their career as well, right?  A player is still providing positive value to his team as long as he’s above replacement, even if he’s below average, so he should still get credited with positive value rather than negative to his career total when we look back at how good different players careers are, right?

WAR
= Value
= “how much would you pay the guy for that performance?”

Well, I’ve got nothing to add to the ongoing discussion.  I just want to comment on WAR. 

What I like about WAR is that it puts responsibility into the hands of the fans.  We could be lazy before.  We could be spoonfed VORP and EqA, statistics we had been told were very good, yet which had almost mystical calculations. Now, the dialogue is far more open.  Here’s what WAR is, and you cannot really talk about it if you do not know it.  Those who implement some kind of WAR are talking at length about the weights they’re using and why, explaining and justifying their formulas.  With VORP, it was just “Here’s a number, and it’ll help you win forum wars.”

When I had tried to explain VORP to my non-sabermetric friends several years back, I had trouble answering their questions.  The publicly available information was lacking.  I understood what VORP was trying to do, but outside of the basics I could not explain its methodology.  And many people resist, and rightly so, what they’re told by a black box.  By contrast, WAR’s very easy to understand from the top to the bottom, and all of its pieces fit together nicely.  Why do you use wOBA instead of OPS?  Let me explain that to you, friend.  How do we get a win?  We estimate that one win equals about 10 runs.  Why 10 runs?  Why not 15, 20?  Well, that’s got to do with the pythagorean winning percentage.  What’s it got to do with that? Let me tell you.

In my experience, it invites, and doesn’t kill, discussion.  Of course, there will always be those who quote WAR as a be-all-end-all statistic.  There will be those who look at Ryan Howard’s WAR and then scoff.  Still, it’s better as a tool to opening communication.  I think.  Or maybe we’re all - as in, the non-sabermetricians - just a tad more curious than we were before.

#56, just because the metric says “0” does not mean that a player has no value.  “0” does not mean zero value.  It just means 10 more than -10 and 10 less than +10.  You don’t “need” WAR.  WAR and WAA are exactly the same. You only need to know how many wins less than an average player does a “free” player create.  And you only need to know that for salary purposes.  It might be different for different eras and it might be different for different positions. There are also some “philosophical” considerations when dealing with replacement which I won’t get into.  The concept of “replacement” is not an exact science.