May 7, 2014

The Knuckle Curveball

How it Behaves, And Why it's Becoming So Popular

by Dan Rozenson

Would you be surprised if I told you that over a quarter of the curveballs delivered by major league pitchers in 2013 were thrown with a knuckle curve grip? I certainly was. This was one of the results of the first comprehensive study of knuckle curveballs in the major leagues, which I have conducted after months of data collection. It is limited to PITCHf/x data, but the large population sizes give us a good first look at what knuckle curveballs behave like relative to standard curves.

What is a knuckle curveball?
First, let me define clearly what this study entails. A curveball is a pitch that, when thrown from a conventional arm angle, will drop from its normal trajectory due to topspin. It is also typically thrown slower than other pitches, so it drops significantly from the effect of gravity, as well. The two most important fingers in the curveball grip are usually the middle finger and the thumb. The middle finger and thumb rest on or beside seams on opposite sides of the ball, as in this image. At release, the thumb and middle finger spin the ball forward, giving it the desired topspin. The index finger is not needed at all to throw the pitch, and a handful of pitchers even leave it off the ball entirely.

What I am collectively calling the knuckle curve is really two related grips (although for the sake of convenience I will use the terms interchangeably). The first involves tucking the first knuckle of the index finger up against the ball, like the Cards’ Sam Freeman does. The second, more common variant, is often known as the “spike” grip. Here’s Craig Kimbrel’s spike grip. These two versions of knuckle curveball are thrown by about 75 active major leaguers, according to my count, yet the perception still seems to be that the knuckle curve is something of a novelty pitch.

The perception may have been reinforced by how rarely the “knuckle curve” pitch tag was used in PITCHf/x data. Until last season, only A.J. Burnett, Vin Mazzaro, Chad Gaudin, and Nathan Adcock had pitches classified automatically as knuckle curveballs—and, to my knowledge, only Burnett actually threw one in that bunch. To MLBAM’s credit, they began adding the knuckle curve tag (“KC”) to a number of pitchers in 2013 after asking to see the list I was compiling. They now have about two dozen pitchers whose curveballs are classified as KC. Future analyses of knuckle curveballs using MLBAM’s automatic classifications will thus be much easier.

Player cards on BrooksBaseball.net also now feature grip information for many pitchers, which should increase awareness of the prevalence of the spike/knuckle curve grip.

Methodology
All of the tracked pitch data is from the PITCHf/x system. The pitches labeled “curveball” in this study were manually tagged by Harry Pavlidis. This data (including release point, which will be important later on in this study) has also been corrected for calibration errors in the PITCHf/x system, thanks to hard work by Dan Brooks and Alan Nathan. This is an impressive reduction in systematic error that should lead the reader to interpret the results below with minimal concerns about measurement accuracy.

The population of pitchers who use knuckle and spike curveball grips was generated through a variety of methods. It involved searching for news reports of people with such grips, watching game video, and browsing through thousands of pictures of pitchers’ grips. I have identified over 100 pitchers, active or retired, with such a grip; 84 of them threw a curveball in the 2013 season. In almost every case, I was able to find photographic evidence. Other evidence that I treated with equal weight was a pitcher’s verbal description of his own grip.

The population of the “other” group consists of every pitcher who threw even a single curveball in 2013, but for whom I could not find evidence of a specialty grip. I was able to identify a non-knuckle curveball grip for a majority of the pitchers in this group. There may well be a handful of “false negatives,” but they are disproportionately likely to be pitchers who threw very few curveballs (and thus had fewer available photos of their grip). I am very confident that whatever remaining knuckle curves exist in this population would not substantially change the findings presented.

What the data says
This study will allow some insight for the first time on what has made this pitch grip so popular and what its characteristics are compared to other curveballs. I compared curveballs thrown with either of the knuckle grip variants versus anything else Pitch Info has classified as a curveball in the 2013 season. Here are the physical characteristics of the pitches as we might typically display them on the Brooks Baseball site. The main number in each cell is an aggregate result, and the number in parentheses is the average of each pitcher’s average—the basis for significance testing.

	N	% Total	Speed	H-mov	V-mov	Dragless H-mov	Dragless V-mov + gravity
Knuckle	17805 (212)	25.6%	79.84 (79.48)	2.61 (2.43)	-5.86 (-5.61)	4.31 (3.99)	-48.53 (-48.44)
Other	51876 (150)	74.4%	77.64 (77.11)	3.30 (2.73)	-5.42 (-5.31)	5.42 (4.44)	-50.06 (-50.38)

The most obvious difference in the two groups is that knuckle curves are thrown, on average, more than 2 mph faster than other curves. (This difference is highly significant, with a t value of 6. The vertical movement + gravity variable—likely as a result of the speed differential—was also significant.) Thoughtful readers might ask whether knuckle curve throwers simply throw harder in general. As it turns out, the aggregate fastball speed of the knuckle group was 92.75 mph, and the other group averaged 93.09 mph. That suggests that the knuckle grip itself has an effect of adding 2.5 mph to a pitcher’s curveball. Additionally, the spin data suggests that rotation rate on knuckle curves is roughly equivalent. The spike and knuckle grip group was therefore able to add speed to the pitch without sacrificing spin deflection.

A sabermetric theory for the knuckle curve’s appeal
A pitch thrown faster has less time to drop due to gravity and drag. For a pitch that travels slowly, as most curveballs do, that means that pitchers need to begin the arc of the pitch higher than they do for their other pitches in order to have the ball cross the plate in the strike zone. Astute batters are often able to spot curveballs directly out of a pitcher’s hand for this reason. It stands to reason that using the knuckle grip can negate some of this advantage for hitters, if a curveball with similar spin deflection can be thrown faster.

To test this, I had the ever-helpful Harry Pavlidis draw up a diagram showing the average pitch trajectories of the fastballs and curveballs thrown by right-handed pitchers to right-handed batters in the 2013 season, separated by whether they were knuckle curve throwers or non-knuckle curve throwers. The curveball characteristics of RHP to RHH mirror the overall trends: on average, the knuckle curves were thrown 2.4 mph faster, and their drop due to gravity, drag, and spin deflection combined was slightly less.

To begin with, here is the side view of the fastballs. Knuckle-curve throwers are in orange, with non-knuckle throwers in black. The scale is shortened somewhat to make differences in the flight paths more visible.

No real difference. Now for the curveballs.

Two things jump out. The first is that the orange dots are spaced farther apart, which reflects the faster speed of the knuckle curves. The second is that the balls are released noticeably farther apart in height and maintain separation throughout the time that hitters would be trying to pick up the ball’s pitch type, though that gap closes as the ball arrives at home plate. In other words, it appears that pitchers can release curves with the knuckle grip from a height and angle that’s closer to their other pitches than those throwing a conventional curveball can. The combination of faster speed and subtler trajectory could create a more effective offering for some pitchers.

I tested both the overall “break” in trajectory for the two types of curveballs and the overall distance (on the x and z axes, with y set at 55 feet from home plate) in release points between fastballs and curveballs for both groups of pitchers. Both tests showed small but statistically significant differences—about half an inch of extra “hump” for non-spike curves and a release point a half-inch closer to the fastball release point.

Now let’s look at how knuckle curves and regular curves fared against hitters last season, at the aggregate level.

	GB%	PU%	HR/BIP%	HR%	BABIP	Swing%	Whiff/swing%	AVG	SLG	ISO
Knuckle	56.2%	7.4%	3.0%	0.44%	.301	39.5%	31.2%	.215	.326	.111
Other	49.6%	11.8%	3.0%	0.46%	.279	38.6%	29.6%	.211	.322	.111

There are a couple of trends in the data presented above. There is a significant tendency toward ground balls and away from pop-ups among knuckle curve throwers, which perhaps explains the higher BABIP. (The GB rate difference was significant at the 99 percent level). However, there may also be a tendency to swing and miss more frequently (which fell just short of being significant at the 95 percent level). These trends do not prove, but are consistent with, poorer pitch recognition by the batter (on average) when a knuckle curveball is thrown. PITCHf/x data cannot verify that the impaired pitch recognition explanation is true, so future research will have to address it. Another factor beyond the reach of PITCHf/x that may explain the ability of pitchers to disguise their knuckle curveballs is arm speed more consistent with a fastball. I encourage statheads and scouts to collaborate on investigating possible links in that area, too.

A pitcher’s explanation
To get a pitcher’s perspective on why the knuckle curve grips have become so popular, I consulted second-generation starting pitcher Brian Bannister. Bannister threw a curveball with a conventional grip, but he was also a teammate of spike curve maestro Gil Meche. He’s also fluent in sabermetrics and PITCHf/x, and he looked at a draft of this study to feed me some ideas. Bannister took me through some pitching theory and in-game experience that could explain why the alternative grips have reached so many players.

The first reason, naturally, is that a pitcher must have a “feel” for a pitch he’s throwing. Bannister said, “Some guys also like the feel of having a finger on top of the ball, like they’re not going to lose it. It’s kind of a confidence thing.” That makes sense.

In terms of pitching theory, Bannister described a philosophical approach to the curveball that he said he and Rick Peterson share, among others. I’ll let Bannister explain:

There’s actually two ranges you should throw a curveball in, and they get better as you get more towards one extreme or the other. You either want to throw a curveball really hard, if you have the velocity and the spin, because that extra spin is really going to make it break faster—or, if you don’t have the velocity and the spin, you actually want to throw it slower, because then you have the force of gravity to help the ball break. So, the more time you actually give the pitch, the faster it’s going to break into the zone. So there’s kinda this gray area in the middle where you really don’t want to be, and if you throw your curveball around the mid- to high 70s range, it generally isn’t successful at the big league level. You either want to be at least 80 miles an hour plus, or you want to be almost 72 to 73 or slower. [With the slow curve] It’s really about how much sink it has when it’s entering the hitting area—or at least, to appear that way to a hitter.

For someone like Zack Greinke, who throws a very slow curveball relative to his fastball, the aim is to disrupt timing and change the eye level of the hitter. But the knuckle curve grips are natural choices for pitchers who want to live on the higher end of the curve spectrum. There is, after all, appeal in throwing a tighter, faster curveball. Bannister thinks it may be a result of the changing strike zone:

I think a lot of guys have tightened up their breaking balls—whether it’s cutters or sliders or knuckle-curves—and they’re really just trying to pound that lower part of the strike zone over the plate, just because you don’t get the calls off the plate like you used to.

That could explain why a grip that used to be associated mostly with power pitchers has been adopted by a wider variety of pitchers—even Jamie Moyer! A curveball that acts a little more like a slider will not break as significantly from the time it crosses the plate to the time it is caught by the catcher. That means umpires and hitters have less reaction time, but also less movement to adjust to. Greinke himself said recently, “My curve is so slow, that they might say ball, and if not, they can still react to it. … [Justin] Verlander’s is so sharp that by the time they realize it’s a strike, they can’t swing at it.”

Conclusion
One out of four curveballs in the majors in 2013 was thrown with a “spike” or “knuckle” grip. Pitchers and hitters alike could benefit from understanding the typical characteristics of curves with these grips. PITCHf/x data shows that knuckle curves are thrown faster, without sacrificing spin rate. They also tend to have a release point that better mimics the pitcher’s fastball. For a pitcher like Craig Kimbrel with an explosive fastball, making his curveball look like a fastball out of his hand leaves hitters helpless.

The success of any non-fastball pitch usually depends on the ability of pitchers to make it look like their fastball for as long as possible. My working theory for the knuckle curveball is that its faster speed and lower release cause batters to hit more ground balls and miss more often because they are less able to tell it apart from the pitcher’s fastball. Pitchers who tip their curveballs with slower arm speed or a higher release point might try the knuckle grip to make the pitch more effective. The grip may also be useful for pitchers stuck with looping curveballs that they find hard to control or get consistent strike calls on, or pitchers who want a pitch similar to a slider but without the associated elbow risk.

I owe special thanks to Harry Pavlidis, Dan Brooks, and Alan Nathan for their help with this project.