THE BOOK--Playing The Percentages In Baseball

That is a frightenly bad article.  I have no idea what the author is talking about. Other than the pitches per PA issue, wOBA is wOBA. By definition, two players with the same wOBA have exactly the same value, not counting base running and defense, and not knowing the rest of the team or the lineup slot (IOW, assuming he is surrounded by an average team and he bats in a random slot or perhaps an optimal slot for his profile).

He keeps talking about whether and why two players with the same wOBA might have different values to their teams, irrespective of the “seeing pitches issue.”

1. Acknowledging that there are multiple ways of arriving at the same Weighted On Base Average, would you argue that there is a better way of arriving there?

Other than the “seeing pitches issue, no there isn’t.  wOBA IS THE DEFINITION OF VALUE, again not knowing anything about context.

2. The benefit of taking a walk is increased wear on the opposing pitcher’s arm as well as showing your team more of what the pitcher has that night.

Maybe.  We definitely don’t know.  And if there is a benefit, it is probably small.

On the other hand, the benefit of the high-average, low walk approach is the chance of batting in runs. Is that a fair trade-off?

What?  That is already included in wOBA!  That is why the walk is .72 and the 1B is .9.

3. Aside from the obvious “he does about as well and is a MUCH better fielder and base-runner” arguments, which are admittedly hard to ignore, which would you rather have on your team?

Again, makes no difference, unless you know something about the other players on the team.

4. The real root of all of this is not the value of the walk, but the value of the pitch. Do you agree that a 5-pitch hit is better than a 5-pitch walk (possible batted in run)?

What?  Of course, a hit is better than a walk on the average.

How many pitches does a batter have to take to make a walk more valuable than a hit? What’s the least amount of pitches a batter has to take to make a walk more valuable than a 1-pitch hit?

Again, assuming that we don’t know anything about the outs/base runners, etc., although that is a fairly interesting question, the answer is probably like 20 or more (pitches to make up the difference between .3 and .47 runs - the average difference between a walk and a single).  And the author keeps saying “hit.” I surely hope he means “single.”

I suppose that everything else being equal, you would rather have the high walk guy with the same wOBA since he does see more pitches, but the benefit of seeing more pitches is probably small at best and may not be a benefit at all, as we have discussed before.

---"I suppose that everything else being equal, you would rather have the high walk guy with the same wOBA..”

From everything I’ve seen over the years, I’d rather have the higher HR guy with the same wOBA. (many times, of course, the hi HR guy will also be the hi BB guy.)

"From everything I’ve seen over the years, I’d rather have the higher HR guy with the same wOBA.”

Why?

Wouldn’t that mean that the wOBA undervalues the HR?  Why would it do that?

I am talking about if we knew a player’s true talent wOBA of course.

I suppose that the bonus for getting “lucky” on a HR (meaning there are more runners on bases than average when it is hit) is higher than the one for getting lucky on a single or walk.

I’d rather have the low strikeout guy because he tends to do better in context-dependent measures such as WPA and Clutch. (This does not mean wOBA’s weights are wrong. It just means that wOBA addresses runs and not wins.) I was hoping the article would address the differences between wOBA and WPA/LI among players. Maybe another time.

#5, I don’t think that is true - that a low strikeout guy does better in “wins” even though he does the same in runs.  Tango can probably chime on on this.

I think Jeremy was referring to the “Color of Clutch” articles I did, where I had the fans choose their clutch hitters, who invariably were contact hitters, and the non-clutch where Dunn-type (TTO) hitters, and the so-called clutch hitters might have a bit of benefit.

But, remember, that bit of value they may have would only apply say 10% of the time.

Right, but all else being equal, that 10% of the time is all that matters. Anyway, I just ran a correlation using data from FanGraphs on qualified hitters from 2006-2008 (470 players) and the correlation between K% and Clutch was -.13.

That’s an R^2 of a little over .01 - doesn’t really seem significant.

I saw a markov study recently that showed that a hi hr team (with the same woba) will be more ‘consistent’, thus leading to more wins.

That’s funny dave, as you’ll always hear announcers say that teams that don’t “rely” on the HR are more consistent.

Jeremy: what was the average number of PA in your sample?

One thing to think about, and I don’t think this is what the article was referring to, is that all doubles are not created equal. 

A double by Adam Dunn is obviously crushed an will score a man from first most of the time.  A really slow guy might end up with a single on a hit that would score someone from first.

A Carl Crawford double might also be hit to the wall, but it might be a ball that is in between the fielders and wouldn’t score a person from first. 

Also, grounding out to the right side every time a runner is on second will have a little less negative value than a regular groundout.  Someone from Minnesota trying to impress their manager and get high fives from their team mates might try to ground out to the right side in these situations, and thereby lower his wOBA, but provide a small amount of value for moving the runner up.

I’m sure there are other examples like this, but they are probably close to meaningless overall. The only way, as far as I can tell, for two guys with equal wOBA, fielding, and basrunning to have a significant difference in value would be a sustained difference in clutch value.

Let’s see, the on-base portion of a double is worth roughly .43 runs, and the moving-over portion is worth roughly .34 runs.

Can we break that .34 further?  Let’s see.  Let’s presume there are .3 runners on 1B, .2 runners on 2B, and .1 runner on 3B.

The run value of scoring the guy on 3B is .4 runs for each runner.  To score the guy from 2B, the run value is .6 runs.  And to score the guy from 1B, the run value is .75 runs.  (Rounded for ease.)

Let’ say that the guy from 1B will score 40% of the time (and gets his .75 runs), and the other 60% of the time, he will be stuck at 3B (therefore, he only get .35 runs in those cases).

Let’s add it up:
runner on:
3B: .1*.4=.04
2B: .2*.6=.12

1B: .3*.75*.4=.09
1B: .3*.35*.6=.06

Total = .31

Close enough for our purposes.  Now, the suggestion is that with the runner on 1B, a Dunn double might score the runner on 1B 60% or 70% of the time, and a double from a weak hitter will score that runner from 1B say 20% or 10% of the time.

First off, that’s easy enough to figure.  Just count it.

But, even if you don’t, what’s the impact?  Well, we take this:

1B: .3*.75*.4=.09
1B: .3*.35*.6=.06

And convert it to this for Dunn:

1B: .3*.75*.7=.16
1B: .3*.35*.3=.03

So, in this extreme example, the average hitter gets .15 runs for his doubles, while Dunn get .19 runs for his doubles with a runner on 1B.

However, Dunn is also less likely to score from 2B than Crawford, on a single, isn’t he?

All to say that this is one of those theoretical things that simply doesn’t hold much water.

Even if you want to give Dunn a bonus of say .02 runs for his doubles, and if he has 40 doubles, that’s not even one run.

Much ado about almost nothing.

New dataset including all qualified players since 2000. 1571 players, minimum 486 plate appearances, mean and median both 614, max 778. I correlated five variables with clutch score.

K/PA: Cor. Coefficient = -.10, p<.01
UBB/PA: Cor. Coefficient = -.05, p>.05
HR/PA: Cor. Coefficient = -.17, p<.01
GDP/PA: Cor. Coefficient = -.11, p<.01
BABIP: Cor. Coefficient = .02, p>.05

It’s possible that players the fans love--low strikeout, low homer, low GIDP guys--contribute more to a team’s wins than their run production measures would indicate.

Good job.  Since HR, BB, and K are linked, why not run a correlation with the three of them in one regression equation.  You’ll probably get r=-.20.

What is the actual regression equation that you come up with?

Linear regression equation using the same dataset with plate appearances as my weights.

exClutch = .27 + (-.84*K/PA) + (-.8.71*HR/PA) + (.8*UBB/PA)

Home runs per plate appearance was the only statistically significant independent variable. Adjusted R-squared = .029.

Hmmm… that puts the r at .17, which is the same if you discard BB and K altogether.

By the way, are you regressing against clutch, or clutch per PA?  Because your equation doesn’t seem plausible that it’s clutch per PA, and really, that’s what you need to regress against.

I’m also presuming that:
-.8.71
is
-8.71

My fault using Clutch instead of Clutch/PA. I re-ran the regression, and the results don’t really change.

Yes, that should have read -8.71.

New equation: exClutch/PA = .00044 + (-.0014*K/PA) + (.0011*UBB/PA) + (-.0134*HR/PA)

Ok, that’s better. 

Let’s talk turkey.  The win value of a HR is roughly equal to +.13 wins.  What Jeremy is showing here is that the “clutch” aspect of a HR is worth 10% less, which is fairly significant.

So, if you are a HR hitter, say you hit +25 more HR than the average hitter, this will cost you .3 wins in clutch.

The walks and K numbers are both insignificant, and move in opposite direction.  So, if you get 60 more walks and 50 more K, that cancels out for your clutch score.

The biggest chokers are the high HR, high K, low BB hitters.

I think it’s fairly significant that when I asked Fans to select the clutch hitter that they avoided the HR hitter with big Ks, and instead relied on the contact hitter.  And Jeremy here provides evidence that supports this notion as well.

So, pretty darn insightful from the fans, wouldn’t you say?  It makes sense, to them, why they’d prefer say Jeter to ARod.

Maybe they can’t figure out the extent to which clutch exists, but at least they figured out the direction to which clutch exists.

I love it when the data sees what the fans see.

"Adjusted R-squared = .029.”

“Hmmm… that puts the r at .17, which is the same if you discard BB and K altogether.”

Just want to point out here that he used adjusted r-squared, so that doesn’t mean r is .17.