World Series time! Enjoy Premium-level access to most features through the end of the Series!
November 11, 2010
Pitcher Release Points
Pitcher Release Points
Mechanics are one of the most important parts of the pitching craft, but they have historically been among the most difficult subjects to study with large samples of objective data. Most published work has involved the comparison of a handful of photos or videos of a few pitchers against a subjectively determined standard.
This is not to diminish the value of the scientific study of the subject by researchers such as Dr. Glenn Fleisig of the American Sports Medicine Institute. These researchers have contributed a great deal to our knowledge of pitching kinematics and kinetics, and a number of clubs have taken advantage of this research to some degree. However, lacking the ability to conduct laboratory studies on multitudes of pitchers, the public analysis community has not been able to make much use of the scientific research in the field.
Using PITCHf/x Release Point Data
The emergence of detailed pitch tracking data from the PITCHf/x system in 2007 promised to change that in at least one way. Finally, comprehensive logs of pitcher release point data were available. In practice, however, this data has not been used nearly to the same extent as other PITCHf/x data, such as pitch speed and spin deflection, have been. Several challenges in the data contribute to the lack of widespread application.
First, the publicly available PITCHf/x data cannot tell us exactly at what moment the baseball was released from the pitcher’s hand. The PITCHf/x tracking system captures images from the middle portion of the ball’s flight. It uses those images and the assumption that the ball is under a constant, unchanging acceleration to reconstruct the full trajectory of the ball. Research has shown that the resultant trajectories are quite accurately reconstructed using this constant acceleration assumption. The pitch speed and location reported using this method are reasonably accurate. However, though the trajectory of each pitch can be reconstructed all the way back to the pitcher’s mound, we can’t determine the precise location where the pitcher released the ball.
Nonetheless, PITCHf/x does provide valuable information about the release point. One way to use the information is to simply assume that all pitchers release the baseball at the same fixed distance from home plate. Of course this assumption is not completely true, but it does allow one to determine release points in two of the three spatial dimensions with pretty good accuracy. Since the baseball is traveling mostly in the direction from the mound to home plate, the height and left-right location of the ball does not change much while the ball travels a foot or so toward the plate. The vast majority of pitches have less than one inch of left-right deflection and a quarter inch of vertical deflection in one foot of travel. Thus, an estimate that is within one foot of the pitcher’s actual release distance from the plate can be used to produce very accurate measurements of the release point in the other two dimensions.
Major League Baseball Advanced Media and Sportvision report the release point information in the Gameday application at a release distance of 50 feet from the back point of home plate. They chose this distance so that the release speed would line up closely with speeds reported by radar guns. It is not, however, the distance at which pitchers are actually releasing the ball. My study of pitcher photos and PITCHf/x evidence has indicated that most major league pitchers release the ball around 54-55 feet from home plate. A few very tall pitchers with long strides, such as Chris Young, release the ball as close as 53 feet from the plate. Short pitchers with short strides may release the ball a little over 55 feet from the plate.
Using the reported data at the 50-foot distance is problematic for release point studies. For one thing, the baseball may move left or right by three inches or more between release and the point 50 feet from the plate. Moreover, this deflection is dependent on the pitch type. This can be particularly troublesome in the vertical dimension because pitchers typically release curveballs traveling at a higher initial angle than fastballs in order to keep the faster-dropping curveball in the strike zone as it crosses the plate. Fortunately, the PITCHf/x data set provides enough information to allow calculation of other points anywhere on a pitch trajectory. All release point data in this article was calculated at a release distance of 55 feet from the back of home plate. This distance is certainly not accurate for every pitcher, but it should minimize the error to an acceptable level.
The final problem that has prevented widespread use of the PITCHf/x release point data is the accuracy of the system. Each ballpark has its own pair of PITCHf/x cameras, and Sportvision periodically calibrates these cameras to reference markers on the field. Because of the abundance of good landmarks near home plate and the lack of the same near the pitching mound, the cameras tend to be better calibrated for measurements near home plate. Plate locations are usually accurate to within an inch or so, and plate crossing speeds to within 0.5 mph. Release speeds and release locations have roughly twice as much error as measurements at the plate. When one is looking for changes in release point of two or three inches, measurement errors of similar size are troubling.
If the release point data could be corrected after the fact to minimize the measurement errors, perhaps it could still be useful for evaluating pitchers. Josh Kalk outlined one such correction method in 2007. The correction method used in this article is similar to Josh’s method, with a few changes, most notably that it was applied at a game level rather than a season level. The method produces a substantial improvement in the quality of the release point data.
It is possible that differences in mound height and pitching rubber position between parks may account for some of the measurement errors in release point. The transient nature of the errors suggests, however, that the bulk of the measurement error is due to camera calibration and not to physical differences in the pitching mound. The correction method should work equally well in either case.
Release point data was collected for the fastballs thrown by every pitcher in the 2008-10 seasons and aggregated at the game level. The following charts present data points showing a particular pitcher’s average fastball release point in the vertical and horizontal (left/right) dimensions for each game in which the pitcher appeared.
For example, Brandon Morrow is a right-handed pitcher who played the 2008 and 2009 seasons with the Mariners and 2010 with the Blue Jays. The following chart shows his average fastball release point for each game he pitched in those three seasons.
In 2008, Morrow’s average release point was just over six feet off the ground and little bit more than two feet toward the third-base side of the pitching rubber. In 2009, his release points were closer to 6.5 feet off the ground and closer to two feet from the center of the rubber. In other words, he raised his arm angle slightly in 2009 as compared to 2008. Then, in 2010, he dropped his arm angle, back to a lower slot than even in 2008, resulting in a release point at or just below six feet off the ground and over 2.5 feet from the center of the rubber.
Another way to look at this data and to see the progression of a pitcher’s arm angle and position on the rubber is to look at how the horizontal and vertical dimensions of the release point changed over time.
Each point on the graph represents the average release point in a given dimension for one game. For the first third of the 2008 season, Morrow gradually raised his arm angle. His vertical release point rose from just above 6.0 feet to about 6.3 feet at the same time that his horizontal release point moved closer to the center of the pitching rubber, from about -2.5 feet to -2.3 feet. Consult the first chart to see how that corresponds to a higher arm slot.
Similarly, Morrow dropped his arm angle at the beginning of the 2010 season. His vertical release point moved from near 6.5 feet to just below 6.0 feet, and his horizontal release point moved from around -2.2 feet to about -2.5 feet.
Here are a few other pitchers who changed their arm angles.
Barry Zito dropped his arm angle on May 23, 2008. He maintained a lower arm angle, gradually dropping slightly lower, until he raised his arm slot again on June 1, 2010 and continued gradually raising it until the end of the season.
Carlos Zambrano gradually raised his arm angle throughout the 2009 and 2010 seasons.
Scott Baker’s arm angle was very consistent in 2008 and 2009 but dropped throughout 2010.
Another type of delivery change can be observed in the release point data. Many pitchers shift their position on the pitching rubber over time, and some do so dramatically.
Cole Hamels dropped his arm angle slightly in 2010, but he also moved over a foot farther toward the first base side of the rubber.
Clayton Kershaw gradually shifted over toward first base during the 2009 season.
Fausto Carmona shifted back toward the middle of the rubber from 2009 to 2010.
There was some of both with Justin Verlander. He raised his arm angle in the first half of 2008. Throughout the last half of 2009 and the 2010 season, Verlander gradually moved back toward the middle of the rubber. There are a few other subtle shifts in his data, too.
Release Point Consistency
Which pitchers have the most consistent fastball release points from game to game? Looking at game-to-game consistency apart from the major shifts on the rubber or in arm angle provides a more interesting answer. The values presented below use the root-mean-squared change in release point from game to game with changes over about six inches removed from the calculation.
Here are the pitchers with at least 300 innings in 2008-10 and the most consistent release points from game to game.
And here are the pitchers with the least consistent game-to-game release points.
Here are the release point graphs for a few of the most consistent starting pitchers—Marcum, Romero, and Lester—along with one of the most consistent relief pitchers, Craig Breslow, whose RMS game-to-game change was 1.9 inches.
By contrast, here are the release point graphs for some of the least consistent pitchers game to game. In addition to the worst full-time starting pitcher, Ted Lilly, who had an RMS game-to-game change of 2.7 inches, we show Dontrelle Willis, the worst pitcher in the sample with an RMS change of 3.7 inches. We also show two full-time relievers, Kevin Gregg (2.8 inches) and Kevin Jepsen (2.7 inches).
So far we have discussed fastball release point changes between games, but what about pitchers’ fastball release point consistency within games?
Pitchers who use multiple distinct release points were removed from the preceding table to the extent that the effect was obvious. Bronson Arroyo and Jose Contreras were removed because they used multiple arm angles. Francisco Liriano and Ben Sheets were removed because they shifted on the pitching rubber depending upon the handedness of the opposing batter.
Applications of the Data
Information on release point trends and consistency is good, but what can be done with it? Are certain release points more advantageous than others given a particular pitcher’s stuff? Josh Kalk examined this question in 2008 and did not find much of an effect.
Josh also looked at overall pitch results compared to release point standard deviation and did not find a significant effect. He did not separate game-to-game deviations from in-game deviations. He also did not separate by pitch types or remove pitchers who were using multiple distinct arm angles or positions on the rubber, nor did he correct for PITCHf/x camera problems at a lower level than the ballpark-season. That is not to say that Josh’s method was inferior; he was examining slightly different questions.
Game-to-game fastball release point consistency does show some correlation with the amount of walks a pitcher allows. The following chart shows the percentage of walks allowed by all pitchers with at least 200 innings pitched in 2008-2010 as compared to their game-to-game release point deviations.
The biggest outliers from the regression line come from pitchers who had good game-to-game consistency but poor in-game consistency, or vice versa. However, in-game fastball release point consistency itself did not show any correlation with the amount of walks allowed, nor did the game-to-game and in-game measures combined show a better correlation than the game-to-game measure alone.
It may be fruitful to look at release point consistency for pitch types other than fastball or at consistency between pitch types.
Also, it should be considered that some pitchers may be subtly but purposefully varying their arm angle from pitch to pitch, which might have a more positive result than a pitcher who is attempting but failing to pitch with a repeatable motion. In addition, the ball-strike boundary may be too crude of a measuring stick. Pitching with command to a target would be a better measure of how a pitcher’s release point affects his pitch location. This author proposed a method for measuring command from the PITCHf/x data in The Hardball Times Baseball Annual 2009, and Jeremy Greenhouse expanded upon that concept here and here.
Release point data could be useful in understanding pitcher injury and fatigue. Josh Kalk and Eric Seidman have studied this issue already, but there is more work to be done. Perhaps the improved method for correcting the release point data would help uncover injury or fatigue trends that were only hinted at in Josh and Eric’s previous work.
At the very least, the release point data is interesting in its own right as a description of the approach taken by each pitcher in the major leagues. Who knew that pitchers were changing their deliveries as much as they do?