One Win Curveballs
So far this year Roy Halladay, Adam Wainwright and Javier Vazquez have each provided a win’s worth of value with their curveballs alone. They have saved over ten runs with their curveballs. On the other end of the spectrum is Brad Penny, whose curveball has cost the Red Sox about a win (9.4 runs).
There is a lot that goes into determining the success of a pitch (speed, location, sequencing, delivery), but for a curveball movement is of the utmost importance. The movement of a pitch is how much the spin of the pitch causes it to deviate from a spinless trajectory, and is commonly broken up into horizontal and vertical components. So the vertical movement is how much a pitch sinks or rises compared to expectation based on its velocity, its trajectory and gravity. The horizontal movement is how much a pitch tails horizontally compared to expectation. Positive horizontal movement indicates a pitch that tails away from a RHB and in to a LHB.
Here are the movements of Halladay’s, Wainwright’s and Penny’s curves
(including Vazquez’s would make the graph too cluttered, but his fit in between Halladay’s and Wainwright’s). The gray dots are the movement of all curves for reference.
Halladay’s curve has a wide range of movement, but generally lots of horizontal movement (tailing away from RHBs an average of about 6 inches) and not much vertical break. Wainwright’s curve has lots of vertical and horizontal movement, sinking and tailing away from RHBs by almost 10 inches in each direction. Penny’s curve has little movement what-so-ever (his curves are very close to 0,0).
The two best curves have more movement than the worst one. I wanted to know if generally the more a curveball broke the better. Is it true that ‘flat curveballs’ get crushed? So I found the run value of a curve based on its total movement (a negative run value is better for the pitcher indicating runs saved). The gray lines are the standard errors.
The relationship is quite strong. As the movement of a curve increases so does its success. Not unexpected, but nice to see. Still you cannot predict curve success entirely on movement, there are many pitchers with worse curves that move way more than Halladay’s. But as a general rule the more one moves the better you can expect it to be.
Dave Allen's other baseball work can be found at Baseball Analysts.
More excellent work, Dave. Who has the best left-handed curve?
Wandy Rodriguez, Barry Zito and Clayton Kershaw have saved the most runs with their curves for lefties. They do not have the best on a per curve basis, but on a total runs saved basis.
On a per curve basis Moyer’s is the best for lefties.
Truly a cool piece.
interesting – but it looks like Penny’s curve have better vertical movement than Halladay’s? Also, isn’t it true that backspin will keep the ball in the air longer, and throwing a curve is already creating backspin that the hitter doesn’t have to create? so a bad curve should travel further because of the backspin already imparted on pitched ball.
lasly, I find it humorous that the graphs are based on a pitch with no theoretical spin as a baseline, but knucklers are also charted, and they have no spin, so they are the theoretical baseline!
Yeah Penny’s curve does have more vertical movement than Halladay’s but Halladay’s moves a little more all told (7.3 inches of movement versus 6.6).
Good point about the spinless pitch. I guess I should think about a better way to explain it, since a spinless pitch in the real world will move quite a bit.
oh, I wasn’t dogging your explanation – I just find it humorous. Maybe it’s the theoretical path of the ball without any effects due to an imparted spin (or lack thereof)
First assume a perfectly spherical baseball….
I assume knucklballs move because of air moving over the seams.
@ Doug – I don’t think so, I bet you could get a plain ball to knuckle as well. It’s all in air deflection, which way the pressure releases as the ball pushes through the air – the ball should go the other direction (to the low pressure side).
d
I’m no physicist, but I’m going to disagree with you. A perfectly smooth sphere offers equal resistance in every conceivable direction. In other words, if we had a machine that could blow a thin missile of air at the true center of our perfectly smooth ball, the air would wrap around the ball equally in all directions. The only variables we can add are wind, gravity and the downward plane from the pitcher’s hand to the strike zone, and I doubt even a swirling 30 mph wind could create much knuckling effect on a perfectly smooth sphere traveling at 65 mph.
Contrasting lace shapes on the surface of the baseball creates uneven resistance and pushes the ball in different directions.
you’re right Choo – but we’re not talking about a perfectly smooth sphere – although one could argue they go where they want as well – use a musket ball as an example, they would not be very accurate until a spin was added when they had the ability to machine the inside of the barrel, making them much more accurate because it straightened out the trajectory.
You’re not going to get laminar air flow over a cowhide baseball. Yes, the seams help that, but they don’t impart the movement IMO.
“So far this year Roy Halladay, Adam Wainwright and Javier Vazquez have each provided a win’s worth of value with their curveballs alone.”
What is the comparison point here? Versus if they used their non-curves there? Versus the average pitcher’s curveball? Versus the average pitcher’s average pitch?
Is the runs & wins saved a measure of what happened, or what would be expected to happen?
And why do you think Penny’s curve is the worst, despite it being far closer to the best curves than many others? If you’re using what happened, then I’d guess it’s largely an issue of small sample size.
Anyways just curious, it’s a cool study but I don’t quite understand the inputs.
The value of each pitch is taken as the run expectancy after the pitch minus the run expectancy before the pitch. If you sum the value of all of their curveballs you get over ten runs saved.
I am not sure about Penny’s curve. It is much worse this year than ever before, it could be a small sample size issue, but he has thrown over 250 this year. It could be because his movement is so close the average curve movement.
Could it be the situations in which the curve is used? I’m not terribly familiar with pitch f/x data, so I don’t know if it supports this, but my perception of conventional baseball wisdom is that curves (really, all breaking balls) with large horizontal movement are more effective when the movement takes them away from the batter, and less effective when it moves towards the batter. Hence the proliferation of LOOGY’s with “frisbee” breaking balls.
I’d be interested to see a platoon breakdown of these pitches value. It probably wouldn’t mean much for Penny, but I wouldn’t be terribly surprised if (and these are totally invented example) the value of Wainright’s curve comes almost exclusively against right-handed batters, or Moyer’s from lefties.
Dave,
Good stuff. Three questions:
1) What is total movement? Simple sum of horizontal and vertical movements? Of the sum in quadrature?
2) Any thoughts about the hump in the middle of the plot?
3) Have you looked at run value dependence on horizontal and vertical movement separately? Which is more important?
Thanks,
-John
#2 if I had to take a stab at it, I’d imagine that the bump is pretty much the average break of a typical curveball i.e. pretty much how a batter would expect a curve to break.
1) I use the square root of the sum of horizontal movement squared and vertical movement squared.
2) Like TomG I assume you get that little bump at the average curve movement.
3) I wrote an article at baseball analysts about that. I think that the horizontal movement is important in same-handed at-bats and vertical in opposite-handed.
http://baseballanalysts.com/archives/2009/04/the_breaking_an.php
I’d love to know whose pitch was the one all the way out at 20 inch horizontal!
Unfortunately I think that little blob far off to the right are a mistake. For some ball parks the pitchf/x system was calibrated perfectly. I am working on going back and correcting for it. It was particularily bad last year at the Great American Ballpark and Bronson Arroyo has a curve with lots of horizontal movement. So I think that blob is from Arroyo’s home curveballs.
Here is some evidence that the knuckleball effect can be achieved without seams:
http://www.youtube.com/watch?v=mX0y1kumtss
A soccerball is easily compressed when kicked and is not smooth – it has tiles and stitches.
Great article, thanks! The last bit about high movement / less effective curves made me curious. I’d love to see if you investigate what other variables account for that.
My first thought was accuracy. But I’m not sure what would be the right measure for that. Balls/strikes came to mind, but a good curve might be one that looks enticing but then breaks out of the zone. So maybe O-swing %?
But my second thought was velocity-relative-to-fastball, so that a bigger differential messed up the hitter’s timing more.
Any other independent variables to test out? Again, I’d love to read a follow-up piece that breaks this down even further. Because who can get enough regression analysis?
Re Ryan and his link: What is needed to get a knuckleball effect is a non-smooth ball. For a baseball, the seams serve that purpose. While a soccer ball (as shown in your link) has no seams, it is far from smooth.
Re curve ball movement: I am wondering if location is more relevant than movement. In particular, a hanging curveball might have lots of movement but will get crushed if left high in the zone. I recall having looked at hanging curveballs a year or so ago and found that the movement on them was not different from better curveballs but the location was high. I don’t recall now if there was a correlation with release point or perhaps initial velocity direction.
That is really interesting, I will have to look at that. Maybe recreate my one plot but with vertical location rather than movement.
Maybe this is an ignorant comment, but I wonder what effect a pitcher’s other pitches has on their curveball effectiveness. Can we really isolate a pitcher’s pitches like this, without reference to the rest of his arsenal. For example, if Roy Halladay has a really great fastball, does that make his curveball better, even if it is not “objectively” better (as measured by break, velocity, etc.)?
And like the last commenter, I also wonder about location. From subjective observation, when Brad Penny is pitching well, he keeps his fastball up in the zone and his curveball down, but when he throws his curveball up in the zone, it usually gets crushed.
Being a Newbie, I am often searching on-line for articles that may help me. Thank you
With all the stuff I am getting confused at the moment.