THE BOOK--Playing The Percentages In Baseball

Why do you think BABIP is so low at 2-strike counts?  As I see it, it could be a number of things:

1) fielders being more “awake” and ready for a ball in play, knowing the hitter is on the ropes and likely “protecting the zone”

2) hitters “protecting the zone” and swinging at pitches they can’t make as solid contact with as they could on a 2-0 pitch they decide to swing at

3) biased sample, in that pitchers who reach 2-strike counts are somehow better at inducing outs on BIP (I swore I read that there was some slight correlation here)

Love the analogy that having a Pujols in your lineup is like having an average hitter starting at a 2-0 count. That’s an elegant analysis of the amazing skill hitters like Pujols have.

Mike, #2 for sure. This has been my criticism of the hitting philosophy of the Milwaukee Brewers for the last 3 years under defensive hitting coach Butch Wynegar. I believe that, with two strikes, increasing your risk for a swing and miss, or caught looking, in exchange for increasing your potential of hitting the ball sharply is almost always worth it.

In other words, I would hypothesize that increasing your potential to put the ball in play (widening the strike zone, shortening your swing) thereby decreasing your chance of not making an out is rarely worth it.

I think #1 and #3 in Mike’s question are either non-factors or extremely close to non-factors. #2 is pretty much it, but it’s not merely an issue of “protecting the zone” (i.e., swinging at marginal strikes) so much as it is also an issue of the batter simply not being in a position to wait for a better strike to swing at. Batters with 0 or 1 strikes can and do lay off strikes that aren’t what they are looking for, but this is not an option with 2 strikes.

If David Appleman is out there, this can be answered by looking at BABIP by the 9 zones.  Presumably, we will find:

1 - That the BABIP for a particular zone is the same, on a 0-2 count as it is for a 3-1 count.

2 - The distribution of such balls in play are disproportionately larger on outside pitches for 0-2 and inside pitches for 3-1

That is, the entire difference in BABIP between the 3-1 and 0-2 count is simply that the batter sees better pitches on 3-1.

I remember there being some debate in the wake of Moneyball as to “hinge counts” ... someone in the book had said that 1-1 was the most important count for a pitcher, if I recall correctly, due to the (perceived) difference in pitching with a 2-1 count instead of a 1-2 count.  I don’t have the book in front of me, so I cannot recall the exact claim.

If your count is 1-1, a ball gets you to 2-1 (a .372 wOBA passing through 2-1) and a strike gets you to 1-2 (though which passes a .237 wOBA), so that’s a swing of .135 points of wOBA on that count.  (I assume it is proper to use the “through” figures for this instead of the “at” figures.)

Anyway, if you do that for every count, the swings are:

3-2 1.00
3-1 .597
3-0 .510
2-2 .403
1-2 .290
0-2 .237
1-0 .206
2-1 .200
2-0 .198
1-1 .135
0-1 .102
0-0 .088

Obviously two-strike and three-ball counts are “high leverage” than early counts.  I’m not sure how that informs the discussion of what count is most “important”, and of course what may be important as a teaching tool from a pitching coach to a pitcher may not be the same as what is important in terms of results.

Hawk,

You may find interest in Craig’s article, which is similar to what you did:
http://www.hardballtimes.com/main/article/the-importance-of-strike-one-part-two/

Not sure how you got 3-2 as 1.000.  Don’t forget that in wOBA, a walk is 0.72.

When you compare to Craig, just remember that runs-delta per PA x 1.15 is wOBA-delta.

Looking at the BABIPs, is there some selection bias among hitters? Presumably Pujols is more likely to get a 3-0 count than Neifi, and it goes the other way for 0-2 counts.

I actually don’t have the wOBA formula in front of me, so I was just kind of guessing at the 1.000.  So that would change the figures for the three-ball counts.

http://www.insidethebook.com/woba.shtml

***

I’m sure there’s a selection bias, but how much can there be?  Here’s Pujols and Everett:
http://sports.espn.go.com/mlb/players/splits?statsId=6619&type=batting&year=2006

http://sports.espn.go.com/mlb/players/splits?statsId=6389&type=batting&year=2006

Pujols went to 3-0, 3-1 92 times.  He went to 0-2, 1-2, 2-2 182 times.

Everett was 36 and 203.

Those hitter counts mimic their number of walks.

As a shortcut, I suppose you can simply do:
sum(wOBA*walks)/sum(walks) for every player
That’ll give you the talent level at hitter’s counts.  And replace BB by K for pitcher’s counts.  Repeat the same for pitchers.

My guess is that you’ll see something like a .350 hitter using the walk formula and .330 using the K formula.

And when it comes to BABIP, it’s probably even tighter to non-existant.

This is minor, as the error is small, but I thought I would point out that the stats on the referenced website are not completely consistant. 

(at 0-0 + through 0-1 + through 1-0) does not equal (through 0-0).  AB are different by 24, H by 12, 2B by 2, HR by 1, RBI by 14, BB by 4, IBB by 2, K by 1. 

Not a big deal (BA is .00004 different), but I was playing with the data and noticed that some things didn’t add up right, so I thought I would mention it in case someone else noticed that the numbers did not add up correctly.

I confirm the issue. 

That is kinda weird.  A couple of possibilities for reasons things not to add up:

1 - Since Tom removed pitchers as batters, maybe there was a mid-PA switch.

2 - A batter gets to bat twice, if he’s at the plate and a runner is putout for the third out.  His count is reset back to 0-0 in the next inning.

3 - Players batting out-of-turn (or just about anything else you can think of) can expose a bug in Tom’s program?  As you said, it’s 28 PA out of almost 200,000, so whatever it is, must be kinda weird.

How do the other counts line up?

In terms of the low-BABIP on 2-strike counts, it would be interesting to see how this breaks down with runners on (especially on 2B and/or 3B) vs. bases empty.  With runners on and 2 strikes the hitter wants to at least advance the runners by putting the ball in play, so there’s more incentive to swing at a borderline pitch.  With bases empty, hitters may feel less pressure to swing at a tough pitch.

sum(T2-0,@1-0,T1-1,@0-1,T0-2)
- sum(T0-1,T1-0) =

18 AB, 4 H, 0 2B, 1 3B, 0 HR, 6 RBI, 13 BB, 1 IBB, 10 SO

Again, the sum of the results are greater than or equal to the sum of the event(s) they came from.

It seems that Adam Dunn knows what he’s doing on an 0-2 count:
http://greenmen.org/skyking162/2007/08/adam-dunns-performance-by-count/

What sky does is compare Dunn’s performance, relative to the league, and then relative to himself.  Ideally, his double-relative chart should show “1.00” or “100”, since everything’s been indexed.  (We’ll probably quibble as to exactly how to index though… Odds Ratio method might seem more appropriate… and this might change the conclusions.)

On an 0-2 count, Dunn is slighly above his expections across the board.  However, at 0-1, he’s below across the board.  So, you might take exception with how Dunn gets to the 0-2 count (i.e., not approaching the 0-1 count good enough).

There’s tons more that can be done with the data, but I think sky does an excellent job in making it as simple as possible.

The pass-through count data in the main blog entry makes it quite clear what is a hitter’s count: every count with more balls than strikes.

The obvious neutral counts: 0-0, 1-1.

The obvious pitcher’s counts: 0-2, 1-2.

Then the gray area counts: 0-1, 2-2, either both are pitcher’s counts, or both are neutral.

Now, there’s one problem with the data (easily corrected), and another thing we can try.  The problem with the data is that not each player is equally represented in both groups.  Good pitchers are part of the pitcher’s counts more than bad pitchers.  So, if those gray area counts (0-1, 2-2) really look like pitcher’s counts because more good pitchers find themselves in those counts, then they are more likely to be neutral counts.  Like I said, easy enough to correct, if you have the quality of hitters and pitchers for each count.

But, the other thing I wanted to talk about: why do we care about the hitter’s and pitcher’s count?  Well, if it’s a hitter’s count, you’ll get more fastballs.  And if it’s a pitcher’s count, you won’t.  The distribution of the breaking pitches and changeups will also change based on the count you are in.

So, instead of the final performance telling us what we would hope that the batter and pitcher know to be the hitter’s and pitcher’s counts, why don’t we ask the pitchers and hitters themselves?  How often do pitchers throw fastballs in each count, and where do they locate them?  How often do batters take pitches, and how hard do they swing when they don’t?

While I am quite certain that 10 of the 12 counts will give us no surprise as to how the pitchers and batters approach the PA, the two gray-area ones might be a bit more interesting.  Even though the final performance numbers of the 0-1 and 2-2 count are very similar, the approaches by the batter and pitcher might be such as to move one into a pitcher’s count, and the other into a neutral count.

Anyone interested in doing the dirty work?

Joe Sheehan took a crack at it:
http://baseballanalysts.com/archives/2007/11/this_comment_an.php

Fantastic work. Fascinating.  I made a post on his blog.

This suggests one possible theory for the different “path” outcomes discovered by Sal Baxamusa. He found a lower BA/SLG after 1-1 count for 1-0/1-1 than for 0-1/1-1. My theory would be that there’s more information about the balance of power between these two players in the 2nd pitch result, because the pitcher probably threw a FB (59%) yet was still able to get it past the hitter.  The first pitch strike, in contrast, tells us less about this matchup because so many hitters take the first pitch a high % of the time.

I’m putting this link here for my benefit:
http://www.insidethebook.com/ee/index.php/site/comments/pitch_analysis_of_eric_bedard/#8

I’ve got this here thread bookmarked already, and I should have put the contents of post 8 and 9 of the above thread into this here thread.

Carry on…

Some great data by Mike Fast:
http://mvn.com/mlb-stats/2008/02/18/breakdown-of-balls-in-play-by-count/

bump

I’ll highlight this thread for those who missed it in the past…

oh… so much work, very appreciate! thanks

Am I reading this right? A guy who gets to 3-0, and then the next pitch is the last of the at-bat, has a 95% chance of getting on base?

If he swings and puts the ball in play, he has a 34% of it being a hit.

If he swings and misses or fouls it at off (through 3-1), his wOBA drops to .490.

Also, does the roughly 5,000 PA difference between “at 3-0” and “through 3-0” mean that on roughly 3 out of every eight (or 38.2%) plate appearances, the pitch is taken and Ball 4 is called?

The first thing and the second thing together (again, I might be reading this wrong), if you swing on 3-0 and put the ball in play, you have a 34.3% of getting a hit, but if you take, you have a 38.2% of taking a walk on that pitch?

Am I missing a parsing of stats somewhere that shows the the wOBA for those who swing on 3-0 (in play or not) and those who take 3-0?

"Am I reading this right? A guy who gets to 3-0, and then the next pitch is the last of the at-bat, has a 95% chance of getting on base? “

No, you are not reading it right.  Well, you are reading it right, but not interpreting right.

You are saying the following: “I know that he’s at a 3-0 count.  I know that the atbat ends on the next pitch.  What happens?”

First off, seeing that a batter is at 3-0 means that he is taking over 90% of the time.  Seeing that the batter is at 3-0 AND YOU KNOW THE NEXT PITCH IS THE LAST PITCH means that he is taking even more.

You know that if he swings AND THE AT BAT ENDS HERE (meaning it is not a swing and miss, or foul ball), he’ll get a hit 40% of the time.

So, if you figure that if you have your two preconditions:
- batter is at 3-0
- next pitch is the last pitch

It’s pretty darn easy to see an OBP of .950.

Indeed, if a batter takes on 3-0, and even though it’s a called ball 35% of the time, BECAUSE YOU KNOW it’s the last pitch, it will be a called ball 100% of the time!

So, he takes say 92% of the time (.92 walks), and the other 8% of the time he swings, he doesn’t miss, and so, get a .400 BA (add .032).  .92+.032 = .952, close enough to what is being reported.

It is critical CRITICAL to understand that you are starting with an endpoint, and inferring results.  You are not starting off with “what would happen if”.  You are starting off with “this is what happened”.  Huge difference.

I apologize for the excessive use of CAPS.  Usually when that happens, it means the writer of the comment is full of it.  This is an exception.

By the way, this is why you almost always want the “through” counts, not the “at” counts.  “At” means you know it’s the last pitch.  And in probability, you never know that.

Bumping…