THE BOOK--Playing The Percentages In Baseball

As someone on the thread at Joe’s blog points out, passer rating is very complex (four inputs, each converted to points on a scale that has both a floor and a ceiling, then are summed and multiplied by some constant), yet is widely accepted.  And just about nobody knows how its calculated, and those who do know that it’s stupid (it weights each component equally, at least in theory and it’s based on passing standards from the early 70s, in which each rate was lower).

What am I getting at?  I believe the easiest path to mainstream acceptance is not through diffusion on blogs and in print by sabermetricians slowly getting picked up by Peter Gammons and eventually by his network, but rather through official acceptance.  Get MLB to print wOBA in the official stats, and you’ll get in the mainstream--no matter how complex it is or whether anyone understands it at all.

Of course that’s easier said than done.

If was tasked with picking a stat to push in this way, wOBA would certainly be a fine choice.  I agree with Tango that RC/27 goes too far in its implications, which is a shame because it is a nice counterpart to ERA.  R+/PA would work, since it would include the cost of outs but not lead to incorrect applications like R/O.  Of course, wOBA is R+/PA with rejiggered weights, more or less.

I’ve felt for a long time the slash line conveyed a ton of information.  If we see .300/.350/.550 against .300/.400/.450, we know a lot about those two batters.  The first is a hacker who makes great contact with power, maybe a Vlad Guerrero type.  The second is a singles/doubles hitter with a great eye for the strike zone who draws a ton of walks, a Wade Boggs type.  It doesn’t do much for ranking players, but it does a great job of describing them in a short space.

I’d like to know how Poz gets through that whole post, and the comments, and even says he’s read “The Book,” and still manages to not mention wOBA.

I kinda still think he wants a Bill James stat to be the right one. Sorta like uzing the Bill James baserunning numbers when EqBRR is freely available, and, unless I’m mistaken, pretty clearly superior.

I’m glad he roots for his friends, I guess.

David, I am sure that K/9IP, BB/9IP, HR/9IP conveys the information we want, but that doesn’t stop us from combining the three into FIP does it?  No one is suggesting supplanting three for one.

The question is: IF I NEED a single number to show, and I have chosen not to show a triple-slash line, what can I do?  Clearly, I would rather show K, BB, HR per 9 IP, but if I have decided to only show one number, then FIP is it.

What, therefore, can you do on the offense side?  OPS and OPS+ convey provably wrong information (to the extent that people use them wrong).  I’d be on board for a linear weights based RC+ metric (analogous to OPS+), if we accept that, like the QB rating, people will take it on faith.

I agree that if MLB.com gets in front on the issue, then that will go a long way toward doing that.

Three options:

1) GPA, or (1.8*OBP+SLG)/4.  Very easy to compute.  Scaled to batting average, but I think that’s a plus.  (Note:  (2*OBP+SLG)/4 would be even easier to compute, and it’s not too bad.)

2) The simplified wOBA that was posted a few months ago:  .36*(2*BB+1.5*H+TB)/(AB+BB)

3) Not a rate stat, but I still like Estimated Runs Produced

I’d love to see wOBA used more regularly but with two caveats.

1. wOBA should be scaled to 1.0 being average. If you’re into baseball stats ebough that you’re comfortable with the scale of OBP, you should be comfortable with a “new” scale. Setting wOBA to OBP in no way makes it easier to understand.

Setting it to 1.0 also makes it very easy to compare players. A player with a 1.2 wOBA is 20% better than average. You can’t make it any easier to compare players. Of course then we would have to change the name.

2. I would like to see wOBA presented both in its pure hitting form and with linear baserunning added. Hopefully there is a good way to account for baserunning that not only includes SB/CS but also the ability to get from first to third on a single, etc. This way we can describe a players hitting ability as well as his overall offensive ability.

I think the key element (beyond Patriot’s excellent point about official acceptance) is finding a scale that is intuitive for most fans.  If you have to stop for a second and ask “is that a good rate?” then it doesn’t work.

Something like the OPS+ scale on which 100 is average (and departures from 100 represent in percentage terms how much above/below the player is) might meet this test.  What if you normalized wOBA to 100?  That might work.  Even better: try to persuade Sean to make that the definition of OPS+. 

Scaling wOBA to the BA scale (.300 = very good) would also be an improvement.  I understand your objection to BA, but the quality of BA as a stat is totally irrelevant for this purpose.  The only question is what scale is readily understood by fans, and fans don’t have nearly as clear a sense of what a good OBP is. 

Something parallel to ERA would be interesting, but then you’re into “what a team of 9 Pujols” would score, which I don’t personally find helpful.

I think that most fans generally accept what MSM (e.g., Fox, ESPN, CBS, TBS) announcers and analysts present to them.  They hear something, it sticks in their head, and they repeat it to other fans who may or may not have already heard the same thing.

Over time, fans have come to trust traditional metrics like batting average and slugging percentage as a representation of player skill, largely because MSM channels/stations/papers/sites have been so reliant on them themselves.  On-base percentage is slowly gathering momentum in MSM and as a result among many MLB fans because it does not threaten the existence of BA and SLG (even though it could eliminate the former, it is not presented that way in the media where it is used).  In this sense, OBP (=OBA) is more of a new entity than it is a modification.

To try to MODIFY a long-established, trusted entity like BA or SLG (or one that is being slowly introduced like OBA) seems like a difficult if not impossible challenge, because it risks confusing people, something that MSM will never want to do.  To me, this means that something like EQA or wOBA is unlikely to ever be adopted.  Similarly, new metrics based on trusted stats like Runs and RBI are also unlikely to stick.  This takes away options denominated in runs like RCG or RC/27 or even VORP.

So what’s left?

[Begin dream sequence.]

For hitters, I actually think that WPA has real potential in the mainstream.  First, it does not threaten any other traditional batting stats.  Second, though NEW, it is denominated in something familiar (Wins), which fans already understand from a team and pitching perspective.  And third, the stat tells a story at the game level and grows to every level above it (series, week, homestand/roadtrip, month, pre/post-break, season, career).

Broadcasters/MSM writers love to say things like “Pujols almost single-handedly won yesterday’s game against these Cubs” ... it makes for great stories about success, failure, heroes, goats, clutch performance, legendary plays, etc.  Imagine telling them that we have the data to back up those stories—and to fuel even more of them!  Under the right circumstances, they’d eat it up, and fans would love them for it.

[End dream sequence.]

I don’t believe there is one “simple”.  Baseball is not a simple game and with simplicity in baseball stats you are going to give up precision.  OPS, RC, wOBA, WAR are about as good as you are going to get.  Can’t have your cake and eat it too.  You can’t order up a baseball stat like you can a big mac thru the drive thru window of McDonalds.
vr, Xei

"Even better: try to persuade Sean to make that the definition of OPS+.  “

In his blog two years ago, I told Sean that RC+ would be preferred.  He disagreed.  My impressions (for whatever it is worth, and based solely on our public discussions here and at his blog) is that Sean has his own ideas, and he’ll stubbornly stick to them.  Then again, one could say the same thing about me.

The only difference is that I think about it way more than he does, which I hope means that I’ve made far more mistakes than he has, and therefore I learned more from them than he did.

***

a) wOBA is linear weights set to the OBP scale.

b) EqA is (effectively) linear weights set to the BA scale.  I have no problem if EqA becomes the standard, as long as Clay fixes a couple of little wrinkles with it.

c) LWR is linear weights set so that 100 is average:
http://tangotiger.net/lwr.html

I did that a good 5 or 6 years ago, if not more.

Really, all Poz (or whoever) had to do is tell us what value should “average player” represent, and we can create a metric in 10 minutes flat.

The above three sets the average baseline to a) OBP scale, b) batting average scale, and c) 100=average scale, respectively.

My preference is LWR.

I also want to remind people why wOBA was created.  I needed to have something that paralleled OBP for study in The Book.  And when wOBA=OBP, then that means that the SLG of the player was proportional to his OBP. 

Finally, the coefficients in the numerator averaged to exactly 1.00 (as it would have to, if you think about it a bit, or a lot).  So, a walk is .7, a single is .9, a D/T is 1.3, and a HR is 2.0.  As you can see, they all spin off from the “1.0” value.  That is, a single is worth less than an “average” time on base (which is part walk, part single, part D/T, and part HR).  A HR is worth double the average time on base.

This is not to say, at all, that wOBA should go mainstream.  It is to say that I am not on board in changing wOBA to wBA in order to get it mainstream.  wOBA serves a very useful technical aspect, and makes sense (in that the denominator is PA, which is the same as OBP).  Creating wBA simply bastardizes a stat that has only one purpose: MSM acceptance.

LWR, while having the potential for MSM acceptance, has its own inherent value of having average=100.  Leverage Index, for example, rose to its level of (niche) popularity precisely because I had the good luck of choosing 1.00 as the average leverage scale.  And, it can easily translate to MSM acceptance.

EqA, to me, is like what wBA would be.  The scale is “cute”, but in a practical sense, useless on its own without serious manipulation.

LWR is a good choice.  Just need to give it a better name....

Are the formulae behind EQA publicly available?  Another reason to like stats like wOBA is that the description on how to calculate it is right out there.  Most won’t care, but it’s available to those who want to look under the hood.

Puck, yes, EqA is public.  All the exact details of the “Davenport Translations” are not, but all EqA really is just (EqR/Out/5)^.4.  EqR is for all intents and purposes a linear weights formula--it’s only non-linearity is in its treatment of stolen base attempts.

Guy: BOR Index?  (BaseOuts Ratio Index) RC Index?  I agree, you need some sort of catchy name.  Batting Index?  I guess this last one works…

I’ve always liked 1.8*OBP + SLG as a simple rate stat for offensive rating.  The average player comes out pretty near 1.000.  The OBP multiplier could be changed to make it exactly 1.000 without any real loss of effectiveness.  My only complaint about it and many other stats mentioned here as possibilities is it treats BB and IBB equally.  Ideally IBB should have 1/3 the value of BB, but then the stat losses its simplicity.

I agree with everyone that says the stat should be scaled to 1 or or 100 as average.  Easiest to remember for everyone. 

I agree with Pinto that no single number can provide as much information about a player’s skill set as the slash line does. 

Ideally, OBP would be figured with IBB subtracted from the numerator and denominator.  1.8 *OBP +SLG normalized to 100 or 1 would be the single rate number player evaluator, but would always be presented with a player’s PAs and slash line.

Peter, right, I also do 1.7x*OBP+SLG such that the league average always comes out to 1.00.

Maybe it’s 1.69 one year and 1.75 the next year, or whatnot.  I can live with that.

The one problem is that the result will be fairly meaningless.  Say it’s 1.7*OBP+SLG and you have a .400 OBP and .600 SLG, so that the value comes out to 1.28.  That number says that the OBP and SLG is each (on average) 28% higher than league average.  But what does it mean beyond that?  If you double that number and subtract 1 (meaning 1.28*2-1), you get 1.56, and that number (loosely) represents the runs created above average index (meaning creating 56% more runs than average).

(Indeed, this is what Forman does with OPS+.)

So, in addition to agreeing on a scale (and for the sake of argument, let’s say that it’s 1.00 or 100), ideally, it would mean something more than just being a number where the more you are up from 100, the better you are.

Do you really need a catchy name?  OPS+ is a terrible name--I thought Production+ was catchier, but even Palmer and company had to give in to the OPS tidal wive. 

I believe the NFL QB rating is actually called passer rating, and the instructions tell you it’s not a QB rating because of all the things it ignores (since it only looks at passing performance).  That doesn’t stop people from calling it QB rating though, nor does the lack of a singular name cause confusion.

I think Batting Index is good enough.

Lwts runs, wOBA, EQA, or GPA, no question.

If you want something that the fans can comprehend how it is computed, OPS, no question.  No question at all.  OPS is so much better than BA or OBP that you have to give its shortcomings a pass.

Now, if you want something that the fans and/or media are going to accept, asking us is pointless. You gotta ask them in some way, shape or form…

One of the underrated aspects of wOBA is how easy it is to explain the conversion to runs to someone.  I was just going through this the other night with an econ professor who is a big baseball fan but not really up to date on sabermetrics, and while wOBA wasn’t necessarily intuitive, it was very helpful that I could tell him that two points of wOBA = 1 run per 600 PA. 

That’s assuming the constant in wOBA is 1.2, so, in years where it’s 1.18 or 1.23 or whatever, that is not precisely true, but it’s really close, and that’s helpful. 

To be able to look at any difference in wOBA between two players and know that I can cut it in half and have a run differential per full season is a huge plus.

I can live with OPS as long as there is no research that quotes OPS.

I have seen the slippery slope of OPS, and it has led us to the lost city of Atlantis.

Yeah, I think the linear weights scaled to 100 is great:  wOBA is a great stat, but people lack the intuitive understanding of on-base percentage that they have of batting average.  It does need a new name, though:  I think linear weights sounds too mathy to be a great name for a mainstream stat.  If counting stats were in the mix, too, I think wRAA is elegant enough to get some level of mainstream acceptance.

OPS is fine with most research, actually, just as is BA and OBP.  As long as the research has nothing to do with predicting runs scored.  If we want to look at the impact that something has on offensive production, which is actually most of the time when we are doing research on batters, then OPS, wOBA, lwts, Base Runs, etc., are all perfectly fine and I doubt that one is infinitesimally better than the other…

I gotta join the group here talking about some form of wOBA scaled to 100.  Or if you want to be silly, wOBA+. 

I consider myself well knowledgeable on the subject compared to most people (not saying that much i know) and yet I have a hard time at times figuring out how much wOBA means.  wOBAs of around 335 or 375 give me issues, for example (I don’t need people to explain these, i get them, it’s just it’d be easier on the 100 scale). 

That’s why Joe likes OPS+ imo....
---------------------------------------
Lets put it this way.  Joe had a column months back about comparing the closers during the 2008 season when he made up a stat called CLOSER+.  It’s described here: http://joeposnanski.com/JoeBlog/2008/11/06/reliever-ratings-2008/.

And I showed it to a friend of mine, who doesnt like sabermetrics generally (he clings to Pitcher wins and losses, RBIs, etc.) and found it incredibly intuitive.  Why?  Because the scale of 100 was REALLY REALLY easy to understand (Also he was a yankee fan, so the conclusion made him happy).

The answer is thus:  A scale of 100 is really really easy to understand.  Setting wOBA to a 100 point scale and then making it easily available would make it much more likely to be mainstream imo than anything else.  The fact that BA is accepted mainstream is more of the fact its been done forever than do to the scale, but 100 scale measures are ridiculously easy to understand and i think would give Joe what he’s looking for.

The problem with OPS is its slippery slope nature.

Once a researcher uses OPS, it gains a certain level of traction.  Now, all the provisions in using OPS doesn’t get carried forward.

Early on in the popularity of OPS, I saw lots of posts about intentionally walking someone, and the commenter would say “1.000 OBP + 0 SLG is 1000 OPS” = IBB is bad.  Then Forman makes OPS+ all popular, and you have lists of best players, even though OPS+ undervalues walks (though not as bad as OPS).  But, all these biases are being ignored, on its way to being a ubiquitous stat.

So, the level of misuse and misunderstanding of OPS and OPS+ is terribly high, which is why I called it a slippery slope stat: it’s a stat that looks and sounds like it’s “all-encompassing”, but it simply is not. 

And, like I said, since only Sean Forman actually calculates OPS+, we have seen that you just need one guy to set the standard.  That’s why, if Forman insists that OPS+ is good (or good enough), then I’d call on Fangraphs or Hardball Times to set the standard for something that is better (while keeping the 100 = average scale).

Ok, if the consensus is on the 100=average scale (like OPS+ or Leverage Index, etc), then the solution is pretty straight forward.

You use Linear Weights Ratio (as noted earlier):
http://tangotiger.net/lwr.html

Divide that number by the league average (typically around .667).

And you get: Batting Index.

Here was an old article of mine that made use of the (then unnamed) Batting Index:
http://tangotiger.net/banner.html

LW and scaling to 100 were also my suggestion, although I think taking a cue from the mentioned passer rating and calling it Batter Rating would sound better.

I was also one who suggested a 100 scale, but I now see a problem:  you can’t have a final rating until a season is over.  This is a real problem:  the average OPS+ in the NL right now is 94! (I assume because Sean uses last/recent year’s data to set the denominator.) I don’t think that approach works for a stat you want widely adapted.  If a player performed at a 120 rate before the All-Star break, that number shouldn’t change come October.

While I (and others here) think something like Batting Index gives a good, concise summary for quickly evaluating a player compared to his peers, there’s one problem which would likely inhibit mainstream adoption. Namely, there’s an annually adjusted variable, and that doesn’t jibe well with how the common baseball fan likes to be able to comprehend a statistic. One could say – perhaps rightly – that this is the typically pointless stubbornness of the baseball traditionalists, but it would be a problem for securing widespread adoption.

As for the name, this is really a marketing and adoption issue, so I would make a couple suggestions with those perspectives in mind.

First, the name should capture the fact that it’s a league-wide comparison. Secondly, the name (and/or acronym) should be easy to say. Third, it should not form an odd-sounding acronym because those have become an instant source of derision (e.g., VORP).

Of the top of my head, my thought would be Player Batter Rating, which also shortens nicely to PBR. There are no technical words there, it is instantly understandable because the name explains itself, and both the name and the acronym roll easily off the tongue.

I don’t know if that’s a problem.  It’s like the escrow issue!

I agree with Ed/8 that WPA or WPA/LI would be a nice stat for the MSM. It’s easy to explain (the total effect of a player’s production on his team’s chance of winning), and, as Harold Reynolds wanted, it takes into account base/out scenarios as well as inning and other situations.

Also, why hasn’t WAR been said?

PBR = Pabst Blue Ribbon.  Try again, please.

You can forget about a “total” player metric.  Remember the history we have: Pete Palmer had the Total Baseball franchise, and is now in charge of the ESPN encyclopedia.  That’s a powerful mover and shaker, and he can’t get his metrics adopted at large.

Even though he invented OPS (he called it PRO for Production) did not make any inroads until… I dunno… somehow OPS came into the vernacular.

So, you have to be aware that there’s roadblocks to introducing something mainstream.

I do believe that 100=average is the way to go.

#28:  I thought the NL as a whole has an average OPS+ of 94 because the baseline for OPS+ is league (or MLB?) player batting with pitchers removed.  Yet the team totals include pitcher batting.

It’s not just roadblocks: there are some genuine limitations to moving to a single metric, chiefly that whenever the numbers for two players being compared are inexplicably close, or inexplicably far apart, people have no choice but to run back to the component numbers we’ve been using all along.

Ever see a car with only one instrument (a mathematical combination of speed, rpm, engine temp., oil pressure, etc.) on the dashboard?

In my mind, I call it Relative OPS instead of OPS+, but I digress…

Have you ever considered Relative wOBA, Tom?  I understand 100 = average right away, but I still have to guess what an average wOBA is.  Off the top of my head, .325 to.335?

I was a fan of Estimated Runs Produced.  It is on the same scale as RBI, Runs Scored, and Runs Created.  But it gives a clearer picture of a players offense than RS and RBI, and you’ve convinced me of the merits of linear equations over additive ones.  But I think you probably have to choose a rate stat over a counting stat if you have to choose one.  Rate stats are more portable.  Not many leagues play 162 games.

"I still have to guess what an average wOBA is”

It is identical to OBP.

As people look for ever finer grained prospective statistics, I think they are ignoring the effectiveness of rougher statistical measures that may not make sense on a one-for-one basis but tell you alot of information in the aggregate.

For example, wins.  Crediting a pitcher for a “win” in an individual game is often nonsense, and there is at least one example of a pitcher pitching nine scoreless innings and not getting a win.  But over a pitcher’s career, the absurdities tend to cancel each other out, and it turns out that the 300 win pitchers really do have amazing careers, either in terms of effectiveness or longvity.

There are too “rough” statistics that I’d like to see.

First, when an out happens in a game, I’d like to see the scorer credit one player with the out.  This would be the player who, in the scorer’s judgement, contributed the most to making the out.  Like with errors, I expect to see tons of mistakes, but over the course of a season the mistakes should cancel out.

Recently I attended a game where I saw three relief pitchers, I will call them A, B, and C, retire the side in three successive innings.  A is a young starter who was converted to a reliever this season and who has had alot of success in the role.  B is a veteran relief pitcher who most people think wind up in the Hall of Fame.  C is a journeyman reliever.

A and B each three ten pitches each inning, and got their batters out with a mixture of strikeouts and a few easily fielded balls.  C got no strikeouts.  One player hit a ball to right field where the right fielder had to run to catch it, one hit a easy grounder to short, and one hit a ball that dribbled up the first base line and the catcher had to run, retrieve it, and just barely made the throw to first.  C used fifteen pitches.

All three pitchers pitched scoreless innings, but if I were assigning outs I would assign three outs each to A and B, one out to C, one out to the right fielder and one out to the catcher.  In aggregate, I think this could tell you alot about a player’s effectiveness on defense.

On offense, I’d suggest assigning effective runs.  A player would get one effective run per home run, one each time he scored from third base, half a run each time he scored from second, half a run each time he batted in someone from second, and one effective run each time he batted in someone from first.  This probably overcredits the effectiveness of doubles, but it has the advantage of being extremely intuitive.

Ed, once you go down that road, you will eventually end up with Linear Weights.  It is IMPOSSIBLE to stop at the point you are suggesting.

Either you make it simple, or you make it complex.  If you do it half-way, no one is buying it.

"It is identical to OBP.”

Understood, but I don’t have the OBP benchmarks ingrained in my mind like I do for batting average.  Maybe it was the convenience of round number benchmarks when I was a kid.  .300 = good, while .200 = Mendoza line.  But I’m slowly moving more towards OBA as people are using wOBA more and more.  (I think Fangraphs uses it now.)

GGC: I would say that before you get to wOBA, you should buy into OBP.

.300 = bad (equ to .235)
.330ish = average (equ to .260 BA)
.400 = great (equ to .315)

I’d like something that is basically Runs per PA, where Runs is based on LW. 

Which I guess is just another request for wOBA set to a different scale.  I know that .330-.335 wOBA is average and .350 is good, but I have a hard time conceptualizing what it really means.  When I see a number that’s on that scale, I expect to be able to say that it represents something on a per-AB or per-PA scale.  AVG, OBP, SLG measure H, times on base, or TB.  It’s a fraction where I can understand what the numerator and denominator are and what they mean.  I can’t tell you what .350 wOBA means per PA or per game or per season.

.350 wOBA = .350 OBP, with proportionate SLG

What about just wOBA converted to runs, like on fangraphs. its very easy to understand, but it doesn’t really have a name. I think wOBA converted to something easy to grasp would work.

Someone suggested WAR, but the problem with WAR and RAR is the definition of replacement, which lots of people have problems with.

Are we talking about “strategically” what would gain acceptance as a popular expression of linear weights as a rate stat?

Well,then I say just keep using wOBA as much as possible on the internet and in conversation, because it can’t be made more simple without making it more “distorting,” and if it’s adjusted any more for parks, leagues, norms, etc., the relative simplicity of the current format (which, however it is scaled, looks like a traditional hitting rate stat). Just be a bit obstinate about it.

The only other suggestion I would have is to simply start using (again, we’re talking the rate stat version) runs created above/below average per 600 or 700 PAs.

I don’t think it matters if the formula is complicated, as people have said using passer rating as a example. If the MLB and MSM supported a complicated formula it would be OK as long as what the stat is measuring is easy to grasp.

I like the idea of converting wOBA to runs.  But to make it a rate stat I would present it as runs per game.  The formula would be 4.25*wRC/PA (4.25 PA is average PA/game).  I’d call something like “runs per game” or “runs produced per game,” abbreviated to RPG or R/G. 

The average is about .500 (.514 this year) which is nice (a .500 team is average).  Great hitters break .900 (Pujols is .934, Mauer .913), but I think it would take an historic season to reach 4 digits.  Willy Taveras grabs the bottom slot at .259.

A few virtues:
1) the units mean something—runs
2) easy to convert to wins—every 100 points means an extra 15 runs or 1.5 wins per season. 
3) Proportional:  a .750 hitter is in fact 50% better at creating runs than a .500 hitter (unlike OPS)

Guy, I was once involved (in an advisory capacity) with MLBAM to create a new stat, and the one they were leaning on was “runs per game”, and using Linear Weights as the engine.  (And 4.3, not 4.25 as the PA… actually, it was whatever was the average.)

It kinda died there.  You need someone there to be a real champion of it.

My main issue is that the average would float every year, unlike say win%.  If people are ok with that, then that scale works well.

The other issues would be league differences, and park impacts.

Tango:  I do think runs per game would make a great uber stat, a big upgrade over OPS.  I don’t think a floating average is a problem for fans, in fact I think it’s probably an advantage as “absolute” stats are much more familiar (BA, OBP, SLG, OPS).  And it would be easy enough for more serious sabermetric types to adjust it for season/league and park differences in their work. 

But of course you’re right that it won’t catch on without a champion and, eventually, some official sanctification.  Maybe poz will lead a campaign if he can be persuaded.....

I really like runs per game.  One would have to decide whether to use park adjusted or non-park adjusted lwts as the “engine.” I would not worry about differences between the leagues though, although I suppose you could adjust for that (it would not be as easy though as a park adjustment).

What I would really like, which is never going to happen, is for the official stat for a player to be his running projection.  That is because the average fan as well as the average commentator ALWAYS uses the current season’s stats, whataver they are, as a proxy for how good a player IS and how good he is going to be tomorrow.  Now, they may start out at the beginning of the season qualifying that, but by the time you get to mid-season or so, EVERYONE uses current season stats as THE number for a player’s true talent.  If I had to choose the ONE thing, and one thing only, that infuriates me the most about mainstream reporting and “analysis,” it is that - using to-date one season stats as a proxy for a player’s true talent.  NO ONE, and I mean no one, but a saberist ever thinks about a player’s projection after mid-season or so.  Occasionally you might hear someone mumble something about a player having a career year or a particularly bad season (like Ortiz) with a slight (ever so slight) implication that maybe he has just been lucky or unlucky so far.

The second thing that infuriates me the most is when people quote how a player is doing “lately” as if that “trend” should or might continue.  As in, yeah, but lately Francouer has been hitting well, so maybe there is some hope for him yet.  There is always hope of course, but the fact that a player has been hitting well (or poorly) LATELY does not change the hope factor very much at all (other than how the “recency” weighting changes the projection slightly).

I realize that the “people don’t really care what the formula is” meme has been beaten to death in this thread, but I can’t help but take another swing at it.  The current go-to stat of this type on B-R, OPS+, is a perfect example.  Of course we all know that OPS+ is OBA/LgOBA + SLG/LgSLG - 1...but what percentage of the users of OPS+ do you think know that?  I’d guess no more than 25%.

One could argue that they *think* they know how it’s calculated (OPS/LgOPS), and just don’t bother investigating further.  Anyway, my t-i-c suggestion was to use the name OPS+ for LWR/LgLWR, and see how many people noticed the difference.

I’m a GPA man myself.  I believe that the BA scale is very intuitive and important to the average baseball fan (I instantly appreciate what GPA means; I have to think about wOBA) and I dislike the 100 scale. It’s like asking people to move from Farenheit to Celsius.  Ain’t gonna happen.

Also, GPA’s easy conversion from OPS makes it easy to understand and accept. I don’t agree that people don’t care what the underlying formula is.  There may be one example in which it worked for football, but that’s just one example. Show me ten and I’ll start to consider it.

Studes: 
I don’t think you can reconcile your two points.  If the formula must be easy to understand, then GPA fails that test as much as the other ideas under consideration. I mean, seriously: you think most fans would find (1.8*OBP+SLG)/4 easy or intuitive? (THT glossary:  “A simple formula for converting GPA to runs is PA*1.356*(GPA^1.77).” I do not think that word (simple) means what you think it means :>).)

I agree that BA is the most intuitive scale.  But I think RPG has stronger advantages.  First, the units—runs—are fairly intuitive, and one thing we want is to get fans to rate offense in terms of runs created.  Second, it’s proportional: .750 really is 50% better than .500.  Third, it actually does convert in a simple way to runs/wins.  Fourth, it’s a bit more accurate (not important for average fans, but if you’re going to go to the trouble of trying to displace OPS, why not get it right?).

You could also have RPG+, normalized to 100 and adjusted for league/park, for those who wanted a more sophisticated measure.  Which is the same as LWR/LWRlg.  The two stats would effectively replace OPS and OPS+.

Studes, what percentage of fans do you think can calculate slugging average?  ERA?  I don’t mean this rhetorically, I’m genuinely curious as to what people think.  My impression is that a lot of casual fans really wouldn’t know how to figure even the most rudimentary statistics.  Admittedly, I might be out to lunch on that point.

I’ve never seen that formula Guy mentioned (GPA^1.77*PA*1.356) before.  An exponent really shouldn’t be necessary to convert 1.8OBA + SLG to runs.  A simple linear conversion should suffice, it’s much simpler, and also it will cause less distortion for extreme players.

I don’t think many fans calculate slugging percentage in their heads, if that’s what you mean. But I do think that most of them (at least, the ones who care) *understand* what it means, mathematically.  And I think they would understand GPA, too.

Plus, GPA is what it is.  My biggest problem with things like RPG is that all the people who hang around here get into discussions about the *best* way to calculate RPG, why one system is better than another, etc.  Just like what always happens with linear weights. IMO, that will always keep stats like that from being accepted in the mainstream.  Too much debate in the background, meaning that stats sites will never have exactly the same RPG or linear weights.  A stat that is accepted by the mainstream must be the same wherever it is displayed, like BA or SLG are.

For researchers, diversity of opinion is a good thing.  For the general baseball fan, not.

I don’t think the passer rating example works for baseball. Passer rating was created because the NFL wanted a better stat to judge quarterbacks, and it was immediately used in its first year of creation as the go-to QB stat (1973). They didn’t have any other all-encompassing stat before then; the year prior, the NFL ranked QB’s based on the inverse ranking of four statistics (completion %, touchdown %, interception %, and yards per attempt).

I’d want a stat that is used in terms of something. GPA is great, but what does it mean? And the same could be said for OPS/OPS+.

Patriot/1: Passer rating is, easily, the dumbest stat in any sport ever. I honestly have no idea how it has caught on in the mainstream, other than football really doesn’t have many stats. Oh, and 100 is “really good,” so that helps.

Why it sucks:
1) It is very complicated. I have a site bookmarked that does the calculations for me. I CAN do them, but it’s not simple at all.

2) The NFL caps the number for no apparent reason. It’s 157.3 or something arbitrary and pointless. The math allows the number to go higher. The NFL doesn’t. It calls this number a “perfect rating.” It’s kinda like if baseball only allowed ERAs to go as low as 1.75, and even if you were better, your number stopped dropping at 1.75. Just because.

3) It doesn’t accurately reflect how good a QB was. Quite simply, the formula is broken. Badly. How so? If you had a fictional quarterback who I’ll call Lyle Morton (not Lyle Mouton), and he threw an incredible 100 passes in a single game, all of them incomplete passes, what do you think his rating would be? 0, right? Negative something? No. His rating would be 39.58. 100 incompletions, no completions, and his rating is 39.58. Let’s take another quarterback and call him Ray Butler. Ray doesn’t throw nearly as many passes as Lyle, and he, too, has a bad game. He throws 35 passes, completes 18 for 160 yards, but he does throw two interceptions. His rating is 40.17. Not even a full point ahead of the guy who, for all we know, spiked the ball 100 straight times.

Sure, the Lyle Morton example is unrealistic. Nobody throws 100 times in a game. Drop it to 50 incompletions. Same rating. 25? Same. 15? Same. 1? SAME. The passer rating formula is so broken that one incomplete pass is as bad as one hundred.

I’d rather listen to Joe Morgan and Tim McCarver talk about grit and hustle for a million years than hear anybody cite Passer Rating as a legitimate stat. I hate it more than anybody could ever hate the intentional walk.

JD, I don’t disagree that passer rating is a terrible stat (this is a quote from my post: “And just about nobody knows how its calculated, and those who do know that it’s stupid").  The point in bringing it up was that it was an example of a statistic that is accepted on faith by many people without any understanding of how it is calculated.  Studes and others have disagreed with the applicability of the comparison to baseball statistics, and they may well be right, but the point was not that passer rating is useful.

I agree with Studes that runs/game or any other linear weight formula would have to establish one permanent value for 1b, 2b, 3b, HR, BB and out, and leave them alone.  No one is going to understand why a double is worth more this year than last year.  (You could adjust them in the “plus” version for us nerds, but not the mainstream stat.) That doesn’t strike me as so hard, but perhaps it is.

I was replying to you, but not intending to disagree. I was more explaining your point through some examples.

I wouldn’t expect any respected sabermetrics person to actually think passer rating is good.

How about batter ERA as figured by OBP*SLG*31. This will get you within .1 of the league ERA.

It would also help if more people were familiar with sabermetric ideas and why they actually make sense. Too many people say things like “I don’t believe in that stuff” or “it doesn’t measure intangibles.”

There are too many different stats. One has to be decided upon and pushed.

As many of you have probably seen, Poz has posted a follow-up with something called “Hittin Average,” which, on a cursory glance, certainly doesn’t seem that much simpler than anything here.

Honestly, from a rhetorical standpoint, if everyone had just said “wOBA!” with a unified voice, that probably would have done the trick, and that could still happen. Just my two cents. It’s not perfect, and I’m not trying to butter anyone else’s bread, but it’s simply, has numbers in a range people can get after a try or two, and really does represent (at least relative) hitter value.

devil_fingers, Hitter Average is just a different name slapped on Tango’s Linear Weights Ratio. Which is a fine stat, I guess (I don’t really get why it needs to have the single at 1 rather than .47, but I mean it works).

The real conceptual hurdle in dealing with wOBA or LWR in comparison to OPS isn’t that the formulas are really any more complex (I find it quicker to compute wOBA from components than I do OPS) but that a lot of the implicit assumptions underpinning OPS become explicit. wOBA or LWR will tell you a double is 1.6 times as valuable compared to a single; OPS will count a double as 1.8 times as valuable as a single but it won’t bother to tell you about it.

I think the disconnect here is over the meaning of “simple.” For almost all of us here, I’d wager that wOBA/LWR/etc. is the simpler option - simpler to use, simpler to fiddle with, simpler to interpret and understand the results. There’s a lot of hideous calculus involved in getting OPS to do what is a bit of rudimentary algebra in wOBA.

The simplicity in OPS is essentially in what it keeps hidden from you. Sometimes I think that BP’s decision to keep their formulas so obscure is what makes their stats so popular - maybe it’s easier to tell a fan “Equivelent Average is like batting average, except it weights everything so a home run is more valuable than a single and so on” when they can’t look at the formula and wonder why a home run is exactly THAT much more valuable than a home run and not some other number.

The comments on the first Poz thread seemed to be a lot more intelligent than the ones on the second.  To wit, post #16 suggests OBA + SLG - BA, with the fact that Ichiro is league average presented as a selling point.  #44 reinvents bases/PA, although at least with the cognizance that its probably been invented before.  There are misguided tangents about hitting with runners on base, averages v. percentages, and other stuff.

Poz’ readers are presumably the kind of people we’re trying to reach--people who are open-minded to statistical analysis but not practitioners themselves.  And yet the effort falls flat, which reminds me why I am usually so quick to proclaim that I have no personal interest in converting people.

As I noted in Poz’s latest thread:
http://www.joeposnanski.com/JoeBlog/2009/09/05/hitting-average/

Really, all one has to do is decide what the user requirements are, and we can create a stat for it. It’s really that simple.

There’s really no point about arguing about which is better, other than to point out the construction errors and limitations.

Patriot/67:

I understand not having a personal interest in converting people. MGL has expressed similar sentiment in the past. And when it comes to fans, I think I agree. If people want to be converted, they will be.

The people who absolutely need to be converted are the people IN baseball. Executives first, then managers. It really doesn’t matter if players are converted because I’m not sure I want them thinking too much while out there anyway. After that, broadcasters (and, I guess, sports writers. But what’s more likely is that the stalwarts die or retire and a new generation of enlightened people does the writing). Once the majority of those people are converted, and despite the McCarvers and Morgans, it’s slowly happening, the fans will eventually follow suit.

Look at how many fans are starting to get the value of OBP just because a lot of broadcasts list it along with AVG/HR/RBI when a player comes to bat. It’s gonna take a while (like with any other non-revolutionary sort of change), but it’s happening.

JD, nobody “needs” to be converted. It’s not like we’re talking about reforming our nation’s health care here - obviously we care about who wins or loses baseball games but that doesn’t mean it matters.

So we try to be correct as much as possible, and to admit error and make correction where we aren’t. And we try to help out anyone that’s interested in learning. And we try to point out errors where they occur.

But we aren’t a religious movement and the point is not to get followers. Speaking for myself, I enjoy doing the research and learning new things - it is certainly gratifying when people appreciate that but it’s not essential.

I like the numbers side of baseball. I really do. But this is the biggest issue I have with the guys who like to go overboard with numbers. There is no single number that should ever be used to rank a baseball player. There never will be one. You can’t properly assign a single number to a guy and have it accurately tell us his value. There are too many aspects to a guy’s game and his value even if you’re ignoring everything other than what happens at the plate.

There isn’t a single stat and we shouldn’t bother trying to find one

Let’s talk briefly about why certain people need to have better “truths” about baseball:

Baseball insiders only from the perspectives of their fans and their owners, because it will help their team win and for less money.  Then again, baseball is a zero sum game for all the teams, so it makes no difference to the league as a whole whether they evaluate players by their horoscopes or using sabermetrics.  To some extent, I guess, the better the evaluation by teams, the better the league is and the more enjoyable it is for the fans.  I am not sure this is true, though.  If fans never saw major league baseball and AAA became the major leagues, would they know any different and would they enjoy it less?

The media, only as it relates to educating the fans.

The fans - does knowing the “truth” enhance enjoyment of the game? I doubt it.  You could even make the argument that it detracts from it.  Why do I think it is important that fans know the “truth” as much as practicable (and that the media portray the “truth")?  Because I think that knowing the truth about anything and everything (other than professional wrestling of course) has a substantial indirect and long term positive effect on the things that actually mean something, like health care.

I cannot teach my kids to be critical thinkers and to search out the truth about things that matter in their lives and the world and then at the same time not care if fans and the media purvey myths and misinformation about baseball.

I’m not sure why Poz wouldn’t just use wOBA.  I mean, it’s simple to calculate, on a familiar scale, and easy to convert to runs above average.  On top of that it’s probably the best formula out there. 

He could take 10 minutes to explain it to his readers (really, you just need a couple of lines), and I’m sure they would understand it.

Also, while wOBA is pretty much the perfect context neutral stat, there is something to be said for quantifying clutch.  Obviously WPA does that fine, but my beef with using it to value players is that is puts too much weight on opportunities. 

My proposition would be for something like xWPA, which would be how many wins a player would contribute with his bat, given a league average distribution of opportunities.  So it would reward clutch, but not opportunities.

As I’ve been reading this thread, I’ve read a lot of excellent points about how the stat should be constructed:

1) It has to mean something in terms of scoring runs or success in at bats.

2) It should be compared to an intuitive “average point”.

3) It shouldn’t modify an existing mainstream stat like BA or OBP.

I think that wOBA is the most accurate linear weights stat out there; however, the “average” part of it isn’t apparent to everyone (as most poeple don’t think on the OBP scale), and it modifies an existing stat.  However, I think it is the best stat out there, and I would like to see it be used to value players. 

So, I am proposing Runs Above Average per 150 Games or per 600 plate appearances, based off of wOBA of course.  That would keep the accuracy of wOBA, while comparing it directly to an average value and it wouldn’t be infringing apon OBP. 

It’s easy to understand and would be perfectly acceptable for formal research.  Plus, it wouldn’t sound to nerdy to say on the TV, and it provides tangible information about players. 

You could also create a variate of that, including a players clutch stats, like I described in #74.

I think that wOBA is the most accurate linear weights stat out there

This statement confuses me. Is this true? Wouldn’t the most accurate linear weights stat vary its weights by run environment?

Not trying to pick on the statement; just wondering if it’s really true.

nick: WPA/LI likely does what you want.  Maybe RE24.

Studes, there is a dynamic version that I posted here, and it is (rougly I guess) the one that Fangraphs uses:
http://tangotiger.net/bdb/lwts_woba_for_bdb.txt

(Page to the right to see the weights.)

So, that might be what Nick is referring to.

Personally, I use .7, .9, 1.3, 2.0 for BB/HBP, 1B/RBOE, 2B/3B, HR, respectively.  When I need precision, I’ll go for it.  Otherwise, I stick with the above.

Thanks, Tango.  But that wouldn’t make it “better” than other linear weight implementations that also vary the weights by run environment, would it?

Apologies if I’m being anal.

I agree with your basic point.  That is, it may not be “better”, but it’s as good as it gets.

If it could be better, I would have made it better!

studes, in every study I’ve done where I put Tango’s wOBA weights in against my “House” weights, the two are practically indistinguishable in accuracy. (And there are a lot of linear weights implementations that do not score as well as either of those, even if they are tuned yearly.)