THE BOOK--Playing The Percentages In Baseball

I’m going to make an authoritative statement so that it leaves no doubt where I stand: you must use the plus 1 method (or differentials).  Anything else that anyone does is almost certainly mathematical gymnastics that gives sabermetrics a black eye and proves that there are damned lies and statistics.

Amen to that.

You know it’s bad out there for OPS+ when half the time it’s mentioned, people start arguing about whether it is biased for the player in question because it underweights OBA.

I recently discovered this post, and today have responded to it at

blog.redlegstats.com/2008/09/28/tango-stumbles-on-abso.aspx

Point taken on the vulgarity.

Point not taken on my impartiality or that I have some conflict of interest.  I’ll hold my honor against anyone out there.

That said, going back to the baseline: choose whatever starting baseline you want, other than .000/.000/.000.  I used something around league average.  You can use Albert Pujols.  Or Neifi Perez.  Take whatever you want as anything that is quasi-reasonable.

Starting with league average is pretty standard.  Starting with Pujols or Neifi is unusual but can be justified.  Starting with zero is all too common, and unfortunately, just plain wrong.

Try to keep the discussion on the merits, not as any kind of personal issue.  I can disparage your work with honest criticism without necessarily making it an attempt of diminishing your character.

I’m not sure I even made the point that you didn’t take.  Maybe it was implied, but the point I was really trying to get at was that I’m not “selling” anything, at least not for money.  Sorry if I made it sound worse than that.

On the merits:  A guy with a league average .269/.333/.430 line has the same 1.75OPS as a guy with .249/.333/.430 line, or one with a .289/.333/.430 line.  Would the derived weights hold true for all three of these players?  What about for players with more or fewer plate appearances?

If I am doing this correctly, using the plus 1 method and your 145-for-540 baseline, 1.75OPS makes a double worth 1.62 times as much as a single, a triple 2.23, and a homer 2.85, while if we use 1 for 4 (a single) as the baseline, a double is worth 1.48, a triple 1.97, and a homer 2.45.  That’s quite a bit of variance in the derived weights for extra-base hits.  Note that I didn’t start with 0/0/0, which you say would be wrong, although I haven’t yet seen you say why.  (You did say I could start with anything other than that.) If we can’t trust the weights for extra-base hits, how can we trust the weights for bases on balls?

I don’t understand about “can’t trust the weights”.  We know what the weights are.  It’s Linear Weights.  A single is worth roughly .47 runs, a double .77, a triple 1.05, HR 1.40, and walk .32 (more or less).  If you divide it all by .47, you get: 1, 1.64, 2.23, 2.98, .68, more or less.

Any metric that is not close to these numbers is not worth figuring, since we have an extremely easy way to figure it (linear weights).

I abhor all things OPS, and the only reason I promote 1.75*OBP+SLG in any shape is as an alternative for those who insist on using some form of OPS.  I’d rather they didn’t even use this better measure, because it leads to problems in extreme cases.

As for batting average, it has little effect once you have OBP and SLG.

You might be interested in this:
http://tangotiger.net/ops.html
http://tangotiger.net/ops2.html

I’d rather they didn’t even use this better measure, because it leads to problems in extreme cases.

When you say extreme cases, do you mean extreme hypothetical cases or extreme cases of actual batting lines for non-pitchers after a non-trivial number of at bats.  I ask because I haven’t noticed any problems using a modified OPS as an estimator for true linear weights.

By adding a constant to adjust to actual runs and then multiplying by a constant to adjust the scale wOBA seems to potentially be altering its estimate of linear weights.  Have you done a plus 1 analysis of wOBA and compared with a plus 1 of 1.75 OBP + SLG for the same data to see which is the better estimator for linear weights?

What I meant by “can’t trust the weights” is that you can’t trust the weights generated by using the plus 1 method to be the same regardless of baseline.  I did not mean you can’t trust the actual linear weights, although I do have some issues with their values shifting over time, and depending on the sample used to generate them.

Thanks for the links.  The bit near the end of part two seemed particularly relevant.  Here are two career lines to consider: Bill Terry’s .341/.393/.506 and Adam Dunn’s .247/.381/.518. Assume these numbers are from the same era/scoring environment, for purposes of illustration.  Each has a career OPS of .899, but we know that a Terry-style .899 has more run-scoring value than a Dunn-style .899, because it is based more on hits, while Dunn’s is based more on walks and extra bases, both of which have less run-scoring value (and lower linear weight values) than singles.  Extreme examples like these are what led me to discard OPS in favor of absolute average.

"you can’t trust the weights generated by using the plus 1 method to be the same regardless of baseline. “

Right, of course.  Different baseline means different results.  The preferred baseline is the league average, since that’s what the typical player will find himself in.  If you wanted to be a stickler, the preferred baseline is 1/30th of the Rays hitting lineup, 1/30th of the Cubs hitting lineup, and so on.

***

Not sure why you have issues with LWTS values shifting over time.  If the environment changes, the impact of each event changes.

***

“Each has a career OPS of .899, but we know that a Terry-style .899 has more run-scoring value than a Dunn-style .899, because it is based more on hits, while Dunn’s is based more on walks and extra bases, both of which have less run-scoring value (and lower linear weight values) than singles.”

How did you conclude that?  Indeed in my link, I make the opposite conclusion.  I show that the player with the LOWER batting average, in the standard hitting environment I chose, generated more runs than the other guy.  If you look at the first and last line in that last chart, you will note these differences for the
.313/.333/.410 hitter and
.185/.333/.410 hitter:

19.7 more HR, 100 more walks, and 119.7 less singles.

The HR are worth 1.4 runs, the walks are worth .32 runs and the singles are worth .48 runs (more or less).  That’s a 57.5 run advantage one way on the singles (higher BA) against 59.5 run advantage the other way on the walks and HR (lower BA).

***

Another way to look at it, for those who don’t want to bother going into details like I do, is consider just OPS.  Most of us agree that OPS is biased against walks.  If you add BA, all you are doing is valuing the walks down even more.  Indeed, some people like to do OBP plus IsolatedPower, which is OBP + (SLG-BA).  Or, OBP+SLG minus BA, the opposite position that ABSO takes!

The truth is that adding BA doesn’t help in OPS, and you shouldn’t do it.

Peter, I’m working on a response for you…

The response to Peter’s question on wOBA has been crafted here:
http://www.insidethebook.com/ee/index.php/site/comments/why_does_17obpslg_make_sense/#33

Any specific question on wOBA should be addressed there, and anything to do with batting average relating to OPS should be addressed in this ABSO thread.

“Each has a career OPS of .899, but we know that a Terry-style .899 has more run-scoring value than a Dunn-style .899, because it is based more on hits, while Dunn’s is based more on walks and extra bases, both of which have less run-scoring value (and lower linear weight values) than singles.”

How did I conclude that?  Because (1) if singles are worth more than walks, an OBP based more on hits than walks is going to be worth more, and (2) if singles are worth more than the extra bases in extra-base hits, a SLG based more on home runs is going to be worth less.

Just to make sure, I calculated wOBA for Dunn and Terry, and Terry’s is indeed higher than Dunn’s, though not by as much as I would have expected.  You might want to check my work in case I made an error in the calculation.

I think OPS is biased against OBP, but that doesn’t necessarily mean it is biased against walks.  I think OBP is biased against hits, because it incorrectly treats walks as equal to hits.  Throwing batting average into the mix helps to correct both biases.

How do you explain my last chart in the second article?

That is very interesting. If you vary either BB or TB alone, the weights and rates all move as I would expect, but if you vary them both simultaneously so as to hold OBP and SLG constant, L-wts and wOBA move counter to ABSO, because you are increasing the number of bases (BB+TB) by more than enough to make up for the lesser weighting of these bases. I’m going to have to think about that for a while.

You listed the run values above as “A single is worth roughly .47 runs, a double .77, a triple 1.05, HR 1.40, and walk .32.” What is the value of an out?  I would like to generate estimated runs by this method and compare to my own run estimator.

You can look at the last line here for one set:
http://tangotiger.net/RE9902event.html

Or here:
http://tangotiger.net/customlwts.html

Sorry it took so long, but the following link is a response to “How do you explain my last chart in the second article?”

http://blog.redlegstats.com/2008/10/18/are-todays-linear-weights-out-of-balance.aspx

There are so many things wrong with this article that I don’t even know where to begin:

http://www.redlegstats.com/ABSOvsOPS.html

Of course ABSO is going to correlate better with RC than OPS, because RC grossly over-values batting average at the expense of walks.

Wow, you divided total average by 3, and then claim to have created a new statistic.

At least you recognize that OPS under-values OBP.

I won’t even comment on this: “Jeff Keppinger and Ryan Freel have hit better than Adam Dunn so far this season.”

In another article, you say that Bill Terry’s 1930 season in which he hit .401 was head and shoulders above Dunn’s career performance. First of all, you can’t compare a single season performance with career performance. I agree with you that Terry’s season was more valueable, than any of Dunn’s most valuable seasons, but probably not by as much as you think. Aren’t you forgetting that the 1930 NL was the highest scoring season during the 20th century. The League RPG for the NL in 1930 was 5.7 RPG and the League Batting Average was .303. Again Terry’s season was more valueable than any of Dunn’s, but you need to be more careful when comparing players across different eras.

My mistake, Bluzer was comparing Terry’s career batting line to Dunn’s career batting line. But you still need to take league context into account when comparing players from different eras.

It may seem like I’m taking an offensive tone to your work Bluzer. After reading your assault on Tango’s character though, it kind of brought out the worst in me. Tango is not trying to sell anything here. In fact, you can browse through the entire book that Tango, MGL, and Andy authored for free on Amazon.com. A few years ago Tango stated that he was living in a one-bedroom space with his wife and kid. Does that sound like someone whose just trying to make a quick buck?

If you are skeptical about Linear Weights and Baseruns, than I highly suggest you read the first chapter of “The Book” or read his 3 part article on “How Runs are Really Created” that he has posted on website.

Bluzer:

You can get the LWTS values for any run environment here:
http://www.tangotiger.net/markov.html

Note that with no outs on base allowed, the run value of each event will be off by a bit.  However, the basic relative changes as you change run environment will hold pretty well.

I am sure that in an environment with few HR that a high BA will be a plus not a minus.

terpsfan101:

On the Terry-Dunn thing:  if the cross-era nature of the comparison bothers you, just think of them as Player A and Player B, playing in the same league.

On Keppinger-Freel vs. Dunn, what I said was true at the time that post was written, which was early in the season.

As for my “assault on Tango’s character,” I think what I said was mild compared to being called, as I was, by implication, a “damned liar” and a “cr@p salesman.” That said, I just think we had a brief flare-up and are both over it now.

Tango, thanks for the link.  I look forward to trying it out.

Just to retort briefly, I had a professional issue with the presentation, and nothing personal.  And I would say the same, whether it was Bluzer, or BPro, or THT, or MGL.  Never should that be stretched to some sort of personal character flaw.  But, in this year of election, it seems that the distinction between a healthy professional disagreement is not always separated from a personal disagreement.

Well, I don’t want to start this up again, but my issue was precisely with _your_ presentation, as well.  It could not have been personal, since I don’t know you.

I am over it and ready to move on, but terpsfan101 brought it up, so I needed to respond to that.

Bluzer, take a look at this chart:

http://tangotiger.net/customlwts.html

This should be pretty similar to what you’re going to get with the Markov values.

The difference between the linear weights value of a single and a home run in a 1 RPG environment is .998. The difference between the linear weights value of a single and a home run in a 10 RPG environment is .835. As run scoring increases, the difference in value between events decreases.

It’s all about having runners on ahead of you, and having people drive you in behind you. In a low run-scoring environment, a home run is a huge deal, even a solo home run, because of how difficult it is for a runner to score when he gets on base.

In a supercharged offensive environment, simply getting on is likely to score you a run, and there are more runners on to drive in.

Just eyeballing it, the biggest difference seems to be in the triple. In a low RPG environment, you’re still almost as likely to be stranded on third as you are to score a run. In a high RPG environment, the triple and the home run have nearly identical values.

The reason you’re seeing the behavior you are has little to do with the LWTS value, anyway, and more to do with how AVG and SLG are calculated.

AVG: H/AB
SLG: TB/AB

In order to keep SLG equal, as H decreases, TB stays the same. So as hits goes down, extra base hits goes up. It’s a mathematical quirk more than anything else, really - batting average and slugging aren’t fundamental measures of baseball, just abstract concepts that we use to stand in for fundamental baseball truths. It’s never good to confuse reality and our models of it.

Colin, that chart presumes the same relative frequency of walks to hits to HR (with just the frequency of the outs varying).

However, Bluzer is talking about 1900s conditions.  If we go here:
http://www.tangotiger.net/markov.html

And change the default value for the HR to 0, we get 3.893 runs per game.  The Linear Weights run value of the single goes up by .02 and the double goes up by .05, while the run value of the walk stays the same.

If we change the HR to 0 and change the number of hits to 11, the run values of each positive event goes up by an extra .04 runs or so.

So, there is a definite change in relative run values among the events, and it’s not always clear the degree (or even direction) to which they move.

It’s very possible that Ty Cobb is being very undervalued under our current equations, simply because we are not considering the run environment “texture” in which he played.  Stealing bases when everyone hits a HR is not helpful.  Stealing bases when everyone only hits singles is very helpful.

Tested this out using the values from the 1907 AL, leaving out strikeouts (those weren’t tabulated, apparently).

http://www.baseball-reference.com/leagues/AL_1907.shtml

The values don’t seem to change that much compared to that chart, at least the relative differences. The Markov values are larger to compensate for the lack of other events.

The Markov isn’t going to tell us anything about the value of the SB, though. And ABSO doesn’t measure SBs, either.

I still stand by my statement that the difference in value between two players of equal SLG but different AVG isn’t measuring anything real, just a “fluke” based upon the inadequacies of measuring offense with AVG and SLG.

Colin:

Note, however, that in the lowest run environment, a walk is worth about 60% of a single, while in the highest it is worth about 70%.  This is what I had in mind as providing a possible correction to the “retrograde motion” in Tango’s chart.

Okay, Bluzer, but none of those have the walk at 30% of the value of the single, which is what ABSO gives us. Using ABSO to try and settle an arguement about the value of the walk is like trying to use kerosene to put out a fire. It’s the furthest thing from the right tool for the job.

What we are really interesting in knowing - and maybe I shouldn’t be speaking for everybody, so I guess I should say what I’m interested in knowing - is what a player’s value is in his appropriate context. If sacrifice hits, bunts and stolen bases are more valuable in certain contexts, we can account for that, and the best way to do that is linear weights. (In fact, neither ABSO or OPS take any of those things into account.)

Well Colin, I never claimed that ABSO was better than linear weights.  I only claimed that it was better than OPS.

I’m not sure where you get the 30% figure, though.  I think it’s considerably higher than that.  We went through the plus 1 method weeks ago, and it makes a great deal of difference what you use as a starting point.

Linear weights have a huge advantage over traditional or quick-and-dirty stats, in that they are self-adjusting.  The values fluctuate, and do not become fixed until the season is over.  They are to some extent like self-fulfilling prophecies because of that.

Let’s simplify AVG, OBP and SLG for a moment, and do them all per PA, rather than per AB for AVG and SLG, like so:

AVG: H/PA
OBP: H+BB/PA
SLG: TB/PA

So that ABSO could be written:

(H+H+BB+TB)/(3*PA)

or

(H+H+H+BB+(TB-H))/(3*PA)

Obviously the real ABSO is a bit more complex than that, due to the mixing of denominators, but that should be enough to give a rough idea of what I’m talking about. I think it’s pretty clear that the construction of ABSO considers a hit to be roughly three times as valuable as a walk.

Looking at the Plus-1 table at the top of this article, and you get the same conclusion:

.17/.53 = 0.321

I really don’t see the reason to go with ABSO in preference to something like wOBA or EqA - I really don’t think you’re gaining any simplicity at all.

I refer the honourable gentleman to the answer I gave some moments ago.

I presume you’re referring to this:

Linear weights have a huge advantage over traditional or quick-and-dirty stats, in that they are self-adjusting.  The values fluctuate, and do not become fixed until the season is over.  They are to some extent like self-fulfilling prophecies because of that.

And - what? The value of an event is determined by its context. That simply is. I don’t know what arguement you’re making here.

No, actually I was referring to the other parts of that post.  The part you quoted was just an afterthought.

You said: “I really don’t see the reason to go with ABSO in preference to something like wOBA or EqA.”

I said: “I never claimed that ABSO was better than linear weights.  I only claimed that it was better than OPS.”

You said: “Looking at the Plus-1 table at the top of this article ... you get the same conclusion”

I said: “We went through the plus 1 method weeks ago, and it makes a great deal of difference what you use as a starting point.”

” (H+H+H+BB+(TB-H))/(3*PA) “

How do you get that a hit would be roughly three times the value of a walk from that, anyway?  Sure, you are adding hits three times, to once for a walk, but you are also subtracting hits once.  Doesn’t that cancel out, so that a hit would be roughly two times the value of a walk, rather than three times?

(H+H+H+BB+2B+2*3B+3*HR)/(3*PA)

Does that work better for you?

And if we’re all agreed that LWTS is better than ABSO AND OPS, why use either?

You’re sneaking in that third occurrence of “hits” as a subset of “total bases”.  Methinks it could be some of those “mathematical gymnastics” that Tango was talking about.

Why use either?  Because, for one thing, you can calculate how well one player did compared to another without having to know what every player in the league did.

H is 1B+2*2B+3*3B+4*HR. TB is 1B+2*2B+3*3B+4*HR. I don’t know how that’s being sneaky.

Or, looked at from that point of view, again simplifying everything as per PA:

AVG = (1B+2B+3B+HR)/PA
OBP = (1B+2B+3B+HR+BB)/PA
SLG = (1B+2*2B+3*3B+4*HR)/PA

(3*1B+4*2B+5*3B+6*HR+BB)/(PA*3)

Or

(1B + 1.33 * 2B + 1.67 * 3B + 2 * HR + .33 * BB) / PA

That’s the intrinsic weights of our simplified ABSO. In reality they’re a little different because of the PA/AB issue.

Colin meant this:
H is 1B+2*2B+3*3B+4*HR.
to be this:
H is 1B+2B+3B+HR

***

ABSO is this according to Colin:
(1B + 1.33 * 2B + 1.67 * 3B + 2 * HR + .33 * BB) / PA

Which you can compare to wOBA as:
(.90*1B + 1.24 * 2B + 1.56 * 3B + 1.95 * HR + .72 * BB) / PA

If we divide my equation by .90, to force the value of the single to 1.00, we get:
.90*(1B + 1.38 * 2B + 1.73 * 3B + 2.16 * HR + .80 * BB) / PA

So, we have pretty good agreement on everything, except the walk which ABSO gives as half the worth it should have.

Now, whether ABSO is better or worse than OPS (which itself is a substandard measure), feel free to argue that.  OPS overvalues HR and better values the walk, compared to ABSO.

Ideally, we stop talking about both, and focus on 1.75*OBP+SLG.  If you are going to alter OPS (which is OBP+SLG), then add 0.75*OBP, and do not add BA.

Doing a similar thing for 1.75OBP+SLG, we get:
1.75 BB
2.75 1B
3.75 2B
4.75 3B
5.75 HR

Divide by 2.75, and you get:
(1B + 1.36 * 2B + 1.73 * 3B + 2.09 * HR + .64 * BB)/PA

Which compares favorably to:
(1B + 1.38 * 2B + 1.73 * 3B + 2.16 * HR + .80 * BB) / PA

As noted, the non-common denominator of the OBP and SLG provides an extra wrinkle.  Nonetheless, you can argue all you want about which is better between ABSO and OPS, but there’s no doubt that 1.75*OBP+SLG is better than either.

And, just so’s we’ve done it, OPS itself:

OBP = (1B+2B+3B+HR+BB)/PA
SLG = (1B+2*2B+3*3B+4*HR)/PA

(2*1B+3*2B+4*3B+5*HR+BB)/(PA)

Or

2*(1B + 1.5 * 2B + 2 * 3B + 2.5 HR + .5 * BB)/PA

It’s a close match for OPS - off about .026 points on average, using all hitters from 2008. OPS is a bit more complex to calculate because of the denominator issue, but nobody’s ever shown me that the extra complexity buys you anything better.

ABSO corrects for the value of extra base hits by tanking the value of the walk; 1.75*OBP+SLG corrects the value of the extra base hit and the walk (while still undercounting the walk a bit.)

If we’re dealing with straight components, it’s easier and more accurate to do this:

2*(1B + 1.38 * 2B + 1.73 * 3B + 2.16 * HR + .80 * BB) / PA

Which is just wOBA fitted to the OPS scale. Call it FauxPS if you will. Fits to OPS nearly as well as our previous OPS estimator.

Qualified starters most underrated by OPS, as per FauxPS:

Ryan Theriot
Kosuke Fukudome
B.J. Upton
Marco Scuataro
Orlando Cabrera

Most overrated:

Chase Utley
Albert Pujols
Alex Rodriguez
Ryan Ludwick
Ryan Braun

Which pretty much matches up with what we knew about OPS already.

If 1.75*OBP+SLG is a good measure, then ABSO should be pretty darn good as well, since AVG is generally about .75 to .8 of OBP.

For instance, this year the NL batting average was .260, while the league OBP was .331.

.260/.331=.785, thus AVG+OBP+SLG is equivalent to 1.785*OBP+SLG for the 2008 NL.

The entire issue I have with ABSO is the incredibly undervaluing of the walk.  That is it, and only it.

Removing the 0.75 from the better 1.75*OBP+SLG and adding in 1.00 for BA removes all the walk value and gives it to the hits.

Do you not agree?

My original post said it all:

event OPS 1.75*OBP+SLG (BA+OBP+SLG)/3
1b 0.45 0.48 0.53
2b 0.83 0.77 0.81
3b 1.22 1.06 1.09
hr 1.60 1.36 1.38
bb 0.23 0.31 0.17
out (0.28) (0.28) (0.28)

And everything I’ve said since is simply a variation of the cold-hearted numbers above.

I have to just take your word for it on the “cold-hearted numbers,” since I can’t reproduce them.  Maybe I just don’t have the background to manage it.  But think about my previous post just on a common-sense level.

You said: “Removing the 0.75 from the better 1.75*OBP+SLG and adding in 1.00 for BA removes all the walk value and gives it to the hits.”

But if BA = 78.5% of OBP, I am not adding in 1.00 for BA; I am adding in .785.

Explain to me, in that case, how BA+OBP+SLG is not the same as 1.785*OBP+SLG.

Because BA is only .785 of OBP for a player with a league average walk rate.

For a player who is an average walker, then yes, ABSO works. For a player with an above-average walk rate, ABSO will underrate him. For a player with a below-average walk rate, ABSO will overrate him. It’s really that simple.

You are including the value of the walk only in OBP - in 1/3rd of the equation - and including the value of the other events in all three terms. Doesn’t it make sense, then, that ABSO would drastically underrate the walk? Because it does.

Is it really that simple?  Is it not possible that as a player’s walk rate increases, the value of the individual walks decrease?  I have a hard time believing that every one of Barry Bonds’ 232 walks in 2004 was as valuable as the walks to the players hitting in front of him—even if you break it down by intentional vs. unintentional walks, and assign a separate value to each of these two events.

Yes, it’s really that simple.

I have to just take your word for it on the “cold-hearted numbers,” since I can’t reproduce them.

Everything I have posted is reproducible (including the Markov program with the source code exposed).  No need to take my word for it.

In any case, your belief in the treatment of the walks is faced against the reality of empirical data.

A walk batting in front of Barry Bonds is more valuable because you have Barry Bonds hitting behind you. [As a larger point, a walk by a leadoff hitter has a different value relative to other events than a walk by the #4 hitter. I published a chart of lineup-specific LWTS on Friday, but MVN’s server upgrade “ate” it. They’re working on it. I can’t tell you how thrilled I am about all of this.]

And yes, walks do tend to occur more often when they are not as valuable, per situation. This is why the value of the hit by pitch is higher than that of the standard walk. Bonds (and Pujols, and a few others) do tend to get pitched around more than the average hitter. For a number of reasons, a linear weights based player evaluation system will start to “fray” a bit at the extremes of performance (although much less than most things deriving from Runs Created).

What you never want to do is to “fix” things at the extremes by breaking them for 9/10ths of your player population. The best you can hope for with a broad-based, linear method of player evaulation - whether it be OPS, ABSO, linear weights, etc. - is to get the tall part of the bell curve right and the extremes reasonably close. ABSO is failing in the tall part of the bell curve when it comes to walks.

There are different ways to interpret the empirical data, and the data are themselves just interpretations of events—either on the part of the umpire or the official scorer.

Are you just going to ignore the idea that as a player’s walk rate increases, the average value of his individual walks decreases?  Consider this:

An IBB isn’t weighted as highly as a NIBB in LWTS, but as you are well aware there are plenty of “unintentional” intentional walks, and even more cases of pitching very carefully, tempting the hitter to get himself out but not caring too much if he happens to walk.  A high walk rate indicates an increased willingness to walk that batter; the IBB is just the most extreme case of that willingness.  Presumably, most of the time, you wouldn’t be willing to walk a batter unless you thought that it was not likely to lead to a run, which would indicate that you believed the LWT value of the walk in that situation was less than 0.5.

Bluzer, did you even look at my list of guys underrated by OPS compared to FauxPS? ABSO has the same problems, but worse.

Now, are teams really pitching around Kosuke Fukudome? Ryan Theriot? Marco Scutaro? You think there are a lot of teams trying to walk Marco Scutaro so they can face Alex Rios?

Colin, Tango said: “For a player who is an average walker, then yes, ABSO works.” If that is true, how can it be “failing in the tall part of the bell curve when it comes to walks”?

If you want to argue that the run value of a walk for an average batter is .32, while it is .35 for someone else, and .28 for a guy who gets a disproportionate number of walks with 1B open, then fine.  That may very well be the case.

But, it is not .17, which is the run value of an IBB!

And in any case, this has nothing to do with the deficiency of ABSO.

As for ABSO working for a guy who is an average walker, this is also true of any metric, however horrible!  For example, let’s say that my formula for RC is:
0.12 * PA

That’s it.  For every single player in the league.  Guess what?  For a guy who is an average hitter, 0.12 * PA is perfectly fine!  It doesn’t work for Pujols or Pena obviously.

In the same way, ABSO does not work at all with guys with lots or very few walks.  I don’t understand why we are talking about this.  There is a world of sabermetrics from A to Z, and this discussion is about the letter A.  There are 25 other things that you can learn to advance yourself on this topic, and a few of them you can learn just by going to my site.  Just the Markov program itself would be the best time you can possibly spend.

Well, I bet you could run a league’s AVG+OBP+SLG and 1.75*OBP+SLG, compare them side-by-side and it wouldn’t make a dime’s worth of difference except for what you guys call the “outliers”—the cases that actually got me thinking about this stuff to begin with.  (Remember the Dunn-Terry comparison?) You couldn’t do that with 0.12*PA.  And linear weights, as I understand them, are derived from team and league data, so what makes you think they would necessarily hold true for individual players, anyway?

But I can see you don’t want to play with me any more, so I’ll just drop it.

I have offered the Markov calculator.  It gives you the weights you are talking about for any kind of run environment.  I already said that the run values of events must change by run environment.  So, run some, and report back results you have found.

***

ABSO has nothing to offer to us, other than the possibility it might be better than OPS, which is a substandard metric.  Once you talk about Linear Weights, you are taking a huge step backward in talking about ABSO.

***

I will have a discussion about linear weights, any place any time.  The issue in this thread is that ABSO seems to tag along for the discussion, when ABSO should simply be dumped to the scrap heap along with OPS, as a matter of professional courtesy.

"ABSO seems to tag along for the discussion, when it should simply be dumped ...”

I must point out that ABSO was the topic of the thread, and you started the thread—about ABSO.

Yes, but once you advance to LWTS, there’s no need to go back to ABSO.

My only point with ABSO is that it severely undervalues walks.  There’s no question about this.  If you can agree to that, we can have a more productive discussion.  If you disagree, then I really have exhausted myself on this topic.  You can count that as me not wanting to play, but I think I’ve played far more than anyone else in their right mind would.

Bluzer, the question really comes down to this. Given the two equations:

(1B + 1.33 * 2B + 1.67 * 3B + 2 * HR + .33 * BB) / PA

(1B + 1.38 * 2B + 1.73 * 3B + 2.16 * HR + .80 * BB) / PA

What reason is there to use the first over the second? You haven’t provided us with anything the former tells us that the latter does not, except for the underrating of the walk.

If ABSO doesn’t add to the discussion - if it doesn’t tell us something TRUE that LWTS does not - then why have it? Why use it?

Re: Colin’s post #47:

I found a cached version of the post that got eaten by your server upgrade:

http://www.google.com/search?rls=en&q=mvn.com/mlb-stats/2008/10/17/setting-the-table/

Thanks, Eli. This is a really frustrating situation right now for all of us at StatSpeak (as I’m sure it is for our readers). I finally got all of my other posts republished last night - that one is still missing.

Why does 1.75*OBP + SLG work? Other than the homerun, it gives perfect linear weights:

0.48 1B
0.77 2B
1.06 3B
1.36 HR
0.31 BB
-.28 AB-H

I’m just looking for an explanation about how or why it works.

http://www.insidethebook.com/ee/index.php/site/comments/why_does_17obpslg_make_sense/

Yes Tango, that thread explained how it worked. I will read it again.

I was hoping for a simple explanation for why it works. I guess it works because OBP and SLG are decent rate stats.

If they had the same denominator, then 1.75 OBP and 1 SLG would give you these weights:
1.75 BB
2.75 1B
3.75 2B
4.75 3B
5.75 HR

Divide it all by 3, and you get:
0.58 BB
0.92 1B
1.25 2B
1.58 3B
1.92 HR

These weights are somewhat similar to wOBA, which we know matches linear weights.

Because of the non-matching denominators, it’s not this simple.  So, this is the simple answer, and my link is the more complete answer.  Hopefully, you will find your answer somewhere between those two responses.

Some people I know are have a discussion on what factor, SLG or OBP, are a better predictor of runs a team scores (no other measures to be used).  Here are the r-squared values of OBP vs Team runs -and- SLG vs Team runs over the last 5 years.

Factor r-squared
SLG .77
OBP .73

Has there been a change that SLG is more important recently or is there something we are missing with OBP?

The reason that SLG might be a better predictor is that there is greater variance at the TEAM level for SLG compared to OBP. 

That does NOT mean you can translate any of that at the player level.

We know EXACTLY the weighting of OBP and SLG, at the player level.  Anything else is obfuscation.  We need enlightening.

Could someone show me how to generate the Linear Weights by adding one additional event to 1.75*OBP + SLG. Tango mentioned something about applying .27*PA, but I don’t understand how this works. I also don’t understand where you apply the -.28 for the out. Here is Tango’s standard batting line.

AB: 540
H: 145
2B: 30
3B: 3
HR: 17
BB: 50
K: 100
HBP: 5
SF: 5