THE BOOK--Playing The Percentages In Baseball

A good choice of baseline might be the prototypical “straight fastball” - the fastball that’s so hittable because it appears to have no movement.  Of course, it has quite a bit of actual movement, but if we “zero” that out, what we’re left with may be a good baseline.

Alan Nathan shared some ideas on what break is here (comment 15)

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

Basically the pfx_x and pfx_z values are most familiar/useful to physicists and are related to the x,z forces on the ball.  The value given in the XML as break_length is more of a player definition of break, and represents the maximum deviation between a straight line from the release point to the front edge of home plate and the actual trajectory of a pitch. 

In the XML, break_length is given just for the y-direction, but Nathan’s website provides several equations that allow you to find the break_length in the x and z directions.  Using this definition a “straight” fastball will have a vertical break around 0, while Rich Hill’s curve will have a vertical break around 14.

More comments here:
http://ballhype.com/story/pitch_identification_tutorial/

***

It seems from what is being said in the above thread is that the break values are meant to relate to spin, but are described in units that describe break (if gravity and/or initial velocity was not a variable, or some such).  Sort of like me saying that Bonds has an On Base Runs Created equivalent of 120 runs, rather than simply saying an OBP of .450.

I suppose, really, what is being said is that the horizontal spin is 20 rotations per sec (rps) and the vertical spin is 30 rps (or whatever), but since no one can possibly relate to that, we’ll try to describe it in different ways, and use this nospin reference.  While the nospin reference is seemingly perfect for the equations set forth, it doesn’t describe what a reader will automatically think it’s describing.  As the old saying goes, “The reader is always right”.

Separating the speed, release point, and spin is fine.  But, it should be done in their own units.  The combination of the three defines the observed horizontal and vertical break.

I don’t like the idea of describing the spin, by itself, in terms of horizontal and vertical break.

I have been arguing for months that the way the “break” is described (the current convention of “versus a no-spin ball") does not allow us to intuitively understand what the pitch appears to be doing.  For example, to say that a changeup rises x inches (because of back-spin), as compared to a no-spin ball, is just a silly way to describe it.

Anyway, only one of the reasons for a typical platoon differential is movement of the ball.  In fact, I am not even sure it is the principle factor.  The other reason is the arm position.  For example, if you are a lefty batter facing a side-arm lefty pitcher, you are going to be bailing out even before the pitcher releases the pitch.

So, yes, it is true that you should be regressing a pitcher’s platoon splits toward the value of a same side pitcher with similar pitches and pitch movements, but NOT toward a pitcher who throws from the other side, regardless of how similar his pitches might be.  So Mariano should NOT be compared to a lefty fastball pitcher because his arm slot is completely different and that also affects the batter’s platoon splits.

In any case, we knew this all the time, and it has always been obvious.  We even talk about it in The Book.  You can’t just regress a pitcher’s platoon splits toward that of all pitchers of the same hand without knowing the arm angle and type of pitches that he throws.  Unless that is all you have of course.  Once you know a little bit about the pitcher (even by watching him throw just one pitch to see the arm angle), you can then “monkey around” with his sample platoon splits.

For example, let’s say that I have a small sample of platoon splits for Chad Bradford and it appears that righty hitters do better than lefty ones.  Well, if I watch him throw just one pitch, I can see that he is likely going to have a huge “normal” platoon split (LHB hitting better than RHB).

So, as we do mention in The Book, and should be intuitively obvious, regressing pitchers’ platoon splits is one of those things where you have to be really careful about what you regress it toward.

All three parameters (speed, release point, spin direction/rotations) are required of course.  I’m not sure of the weights of the three in terms of establishing the regression.  It’s possible that where Mo stands on the rubber (which will affect his release point) along with the spin, and simply throwing one pitch, could put him closer to a LH pitcher than a RH pitcher.

Maybe one of the PITCHf/x stars around here can show various OBP/SLG LHH/RHH splits, by pitches thrown by RHP that are similar to Mariano’s cutter (using all three parameters noted above).

Are the splits excessive for Mariano, or do all pitchers have a reverse platoon split?

Tango,

There are two reference points that the fx guys have come up with - neither are very good nor realistic. The no-spin version is essentially a gyroball, which AFAIK has never been seen in a MLB game. The second straight-line version is also no good. You’d have to put little wings on the ball if you wanted it to fly straight without dropping. No batter has seen that pitch either.

As mentioned already, the best baseline would be an “average” fastaball. Of course, what constitutes average? I can think of two ways to come up with one. The first is to measure a batting practice fastball and use that. If we could have the fx system turned on during the homerun derby we would get a boatload of samples. The other way would be to get an average pitch for all pitched thrown. This would defintely be lean toward the fastballs as many more of them are thrown than any other pitch. I would guess a pitch in the mid-eighties with some (relative to a no spin ball) hop (5”?) and tail (3”?).

with respect to measuring spin on a pitched ball, It can’t be done. The reason you’re getting hop and tail in inches is because that is what the fx system is measuring - not the spin. It’s especially instructive to look at Wakefield’s scatter plot. That knuckler is spinning maybe a quarter turn on the way to the plate but it moves just as much as a spinning pitch does. Spin is really only implied and that implication is not always correct in the case of no-spin pitches.

Regards,

Matt

"The no-spin version is essentially a gyroball, which AFAIK has never been seen in a MLB game.”

Don’t tell CJ Wilson that.

Matt,

I didn’t read the straight line version of break (break_length) to mean a pitch without any drop.  I think it’s comparing the actual trajectory of the pitch to a straight line from release point to where the ball crosses the plate.  The values that you get when calculating the x,z components of break_length are in-line with this.  Rich Hill’s fastball deviates -1.1 inches from the line in the x deviation (in to a LHH) and a 3.4 inch z-deviation.  His curveball has a 3.2 inch x-deviation and a 14.9 inch z-deviation.

If these definitions are just as useful in identifying pitches (which I think they are), maybe the convention can shift to using them instead.  I’m as guilty as anyone of using the pfx values, but as Dr. Nathan indicated, these “new” definitions of break make a lot more intuitive sense for a player.  Hitters (and non-physicist readers) have no idea what a non-spinning pitch would do, but it’s pretty easy to imagine a straight line from the release point to home.

In regards to Wakefield, if you plot a knuckleball he’s thrown using the pfx data it will follow a parabolic course as Matt implies.

I agree with Joe that the straight-line version is from the “release” point to the px and pz values recorded for location at the front of the plate.

Dan,

That is true. The knuckleball goes beyond the basic physics model used in the pitch-f/x system. A knuckler may go left, then right, up, down, whatever. The famous “spinless pitch” is actually a “pitch without spin and a ball without seams, whose average ‘roughness’ is equal to the roughness of a baseball thrown in an average way”. Such a pitch would not knuckle at all.

In other words, the bizarre behavior of the k-ball is not dreamt of in the pitch-f/x philosophy.

More great work from Joe Sheehan:
http://baseballanalysts.com/archives/2007/09/newbreak.php

fyi I posted this at Baseball Analysts. I think there are two markets for this: communication to fans, and sports analysis. Joe’s article is a step forward in the former for sure but I’m not sure about the latter

********

I’ve got to be honest I don’t see what all the fuss is about with regards to the pfx_x/_z values.

In the x direction a pitch with no spin doesn’t move, so the pfx_x value is just the deviation of the ball from the straight line path as it crosses the plate—to me that is reasonably intuitive. In the z direction, for the spinless pitch the only force on the ball is gravity and the pfx_z value is the deviation from this theoretical pitch, as we all know.

The problem I see with break_x and break_z is that it doesn’t always measure true break, in my opinion. The slower the pitch the higher the break (imagine tossing a weak parabolic underarm pitch). This would have a VERY high break under the break_x/_z system but in actual fact won’t break much at all. The parameters (break and speed) are dependent.

Perhaps from communication point of view we should use the break definition but from an analysts point of view I feel it is less meaningful.

Right, it’s a question of clarity.  You have one break that is the “extra break” over and above gravity.  The other break is the break that includes gravity.

In the real world, you want the gravity-based break.

In the analysis world, specifically if you are a pitcher and want to know how much spin you are imparting at release, you want the non-gravity based break.  (You want to know if your changeup and fastball have the same non-gravity influenced spin.) The problem really is using the break to describe this.  Rotations per second, or whatever scale you want, would be more appropriate.  Usurping the gravity-based break word simply causes confusion.

I posted a comment on Baseball Analysts, but this the main point was that the biggest reason for these discussions, which Tango points out, is that there isn’t a consistent definition of break yet.  The parabolic pitch that John mentions is doing something as it comes down, what would you call it?

Despite this confusion though, pfx and break both give very similar results for a given pitch, just with a different baseline.  If you directly compare pfx_z to break z they’re pretty much linearly correlated (pfx_x and break x are even more correlated).

More WOW stuff:
http://baseball.bornbybits.com/plots/players.html

Here’s an accompanying article:
http://www.hardballtimes.com/main/printarticle/a-closer-look-at-jeff-francis-start-against-the-phillies/

***

I also think we should stop using the word break, if we are talking about gravityless movement.  Call it spin movement, as in “the spin on his pitch adds an extra 3 inches vertically, over and above the speed of the pitch and gravity”.  “Break” should be limited to the combination of spin, speed, (maybe release point), and gravity.

Horizontal break makes sense because it is the same as the horizontal spin movement.

I’ve had trouble coming up with a good term for the gravity-less vertical “break”, but vertical spin movement might be the best proposal I’ve heard so far.

Since the common fan has no idea how to infer a “3.5 inch gravity-less vertical spin movement”, why don’t we talk about what it’s really doing?

If we keep the independent variables independent (a. speed, b. spin rotation/direction, c. release point/tangent), and the merging of those variables define the break (or trajectory), then let’s come up with a better unit for spin.

Since the purpose of knowing the spin is solely to know what kind of spin the pitcher imparts on the ball at release (we don’t need to describe it in batter-terms), why don’t we do a “spins per second” or “spins per minute”?  The direction of the spin can be represented by degrees 0 to 360, based on the plane the ball follows along the thumb.  A RH pitcher for example, throwing top spin would basically be throwing a ball at 0 to 90 degrees, and at backspin would be 180 to 270 degrees.

With our 3 terms, we can then describe the pitch as, from the pitcher’s perspective as:
a. released at 96.2 mph

b. at 12.1 spins per second at 60 degrees

c. 3 feet to the right of the mound, 7 feet off the ground, at tangent 10 degrees above the middle of the strike zone (a 60 degree tangent pitch would say be an euphus pitch)

From the batter’s perspective:
a. crossed the plate at 87.7 mph

b. 0.5 feet to the left of the middle of the plate, 3 feet off the ground

c. at an incoming angle of 210 degrees

Wouldn’t this describe exactly what the pitcher is doing and what the batter is seeing?

For the batter, you’d want two angles, horizontal vertical, right?

Doesn’t Mike Fast use spin-rate and spin-angle to classify pitches?

There isn’t two angles… it’s one.  Vertical and horizontal are not angles. 

At the moment that the ball crosses the plate, what is its “angle of crossing”?  Your typical pitch from a RHP would be say south-west (225 degrees), or south-south-west (248 degrees).

I meant angle from a horizontal plane and angle from a vertical plane from hitter to pitcher.

If you just say south-west, you don’t know how extremely south-west it is.  Think of a pitch coming in from due north.  It might be a 50 mph curveball heading straight down or a 100 mph fastball barely breaking down.  But both are coming from the north.

Compass angle plus “radius” would probably be better than horizontal/vertical—like polar instead of orthagonal.

Good stuff.  Right, we want the vector, not the angle.

Re Tango/#17, I think you’re pretty much right on, with a few exceptions/questions.

For the pitcher (c), I have had trouble figuring out how to determine the tangent vector at the release point from the available data (initial position, initial velocity, and acceleration in three dimensions).  I nearly drove myself crazy trying to draw out all the relevant curves and triangles until I finally gave up, so if someone has an idea how to get the tangent vector from the data, I’d love to see it.  It could well be simple and I just missed by trying to represent a 4-dimensional reality on 2-D paper.

For the batter (a), the speed obviously matters, but it seems like it might be the reaction time that matters more than the speed crossing the plate.  In any case, I don’t think we lose much in terms of the physics in most cases by using the initial pitch speed, even for batters.  There’s a lot to be said for statistics have an understandable context to the lay reader, and initial pitch speed has a context much more firmly implanted in the reader’s mind than plate speed.

Yes, the reaction time would be ideal.  We discussed this in another thread.  You could set up your parameters with respect to where the ball is 0.25 seconds prior to it crossing the plate, so that you have reaction distance and reaction trajectory.

(The 0.25 is arbitrary, and could be anything between 0.20 and 0.3 I’d guess.)