Building Metrics for Deception: Release Point Variability

Below is a scatter plot of Brendan McKay’s 2019 release points. What do you see? His release point is slightly higher than most left-handed pitchers and toward the third base side. Combined with a good extension, the Release Score metric rates it an above average release point in deception. 

Then again, that is only utilizing McKay’s average three dimensional release point. Averages can be helpful, but they omit the information of how the data reaches that point. Looking at the scatter plot above, McKay’s release points form a tight cluster. How does that compare to the release clusters of others? What information can we draw from how it does?

The idea of looking at release point spreads has been touched on in a previous BaseballCloud article, ‘BCTips: Release Point Dispersion’. That described the distances between release points within pitch types with the Release Point Dispersion metric. The idea was that pitchers with low dispersions – those with tight clusters of release points – were doing better at repeating mechanics and therefore would have better consistency and command.

One shortcoming of this metric is that it overlooks pitch to pitch interaction. Three pitchers could have identical dispersions but be giving batters completely different looks based on their pitch type’s relative positions. This is not an oversight in the metric, it is just something it was not made to capture.

In creating a metric targeted toward deception, I went a similar route by measuring the distances between release points across all pitches. A low number means the pitcher not only repeats their mechanics, but they release their various pitch types from the same spot as well.

In the first iteration of the metric, Yusmeiro Petit surprised me as a name near the top of the list. He’s a command artist who I had never noted for changing arm slots. Looking at the scatter plots, it was easy to see what went wrong.

Yusmeiro Petit has two distinct release points; not because he drastically changes arm angles, but because he changes his placement on the rubber. Petit almost exclusively pitches outside to batters of both handedness. To align with his targets, he moves toward first base against righties and third base against lefties. Based on where he aligns, batters can tell which release point Petit will release the ball from. Therefore, Petit’s two release clusters are not as deceptive as someone who randomly changes their arm slot.

How can this be fixed? In the first iteration, each pitch was compared to the pitcher’s average release point. So to adjust for players like Petit who change approach by batter handedness, compare each pitch to the pitcher’s average release point against the batter’s handedness.

For most pitchers, this makes little to no difference. Those who it does affect, it affects significantly. Yusmeiro Petit goes from 99th percentile to 34th percentile in release point variability. Similarly, switch pitcher Pat Venditte had his score reduced by a factor over 3.5. 

As much of an improvement adjusting for batter handedness made, it left a lot to be desired. Pat Venditte still has the highest score because of one at bat against switch hitting catcher Tucker Barnhart. With a switch hitter facing a switch pitcher, Venditte was forced to declare he was pitching left handed to which Barnhart responded by hitting right handed. Barnhart knew where Venditte would be releasing the ball, but the left-handed release point was radically different from the right-handed release point Venditte usually attacked right handed batters with. Barnhart could tell what was coming, this methodology could not.

Another data point to note on the scatter plot above is that of Bryan Holaday. No, the Marlins’ catcher was not a master of release point replication, rather he only threw one pitch in 2019. Naturally, that one pitch was equal to his season release point averages yielding a release point variability score of 0. Had he thrown one pitch versus a left handed batter and one pitch versus a right handed batter, subdividing by batter handedness would have driven variability from non-zero to zero. More subdivisions means more players shifting toward Holaday’s position because they do not have enough pitches in each bucket to show variation. This is something to keep in mind as granularity increases.

After adjusting for batter handedness, Keone Kela had the second highest score behind Pat Venditte. Like Yusmeiro Petit had previously, Kela surprised me. He’s a hard throwing fastball-curveball guy with no need to change release points. Further like Petit, Kela featured two release point clusters in 2019 making me question what I missed.

Since the batting handedness adjustment made no difference to his variability score, something else was causing him to change release points. Looking through footage via BaseballSavant, he was changing mound position like Petit, not arm angles. When was he changing mound position?

Prior to his Injured List stint, Kela primarily stood on the third base side of the rubber. He flirted with the first base side for one outing, but promptly switched back. After rehabbing from elbow inflammation, the switch to the first base side of the rubber had been made. Again, the batters knew which cluster of release points Kela would be throwing from thus nullifying the deception benefit.

To fix this issue, the values each release point is being compared to again needs to be redefined. Comparing each pitch to within game averages would solve Kela’s game-to-game changes, but reopen the problem of Petit’s within-game shifts. To fix both, I compared each pitch to the release point of the first pitch of the given at bat. It drove scores toward zero, but also ironed out the wrinkles presented thus far.

This iteration of Release Point Variability shows improvement based on various case studies. The scores of Keone Kela and Yusmeiro Petit (denoted with the red star) regress to the middle of the pack. Additionally, the pitch distributions of Tyler Clippard and Brendan McKay – the two extremes among pitchers with 800+ pitches in 2019 – represent exactly what this metric is trying to capture. 

Clippard has two general release points, but unlike both Petit and Kela, he changes it up within at bats. He’ll throw two seamers and changeups out of one arm slot and four seamers, splitters, and curveballs out of the other. In addition to the generally bimodal release points, he would occasionally let his arm angle slip particularly low. This adds uncertainty for the hitter and disrupts rhythm.

On the other end of the spectrum is Brendan McKay. Having the lowest variability score, he presents with the tightest cluster of release points in the MLB. In this sense, McKay has a consistent, predictable release point and probably mechanics as well. While this sounds like a negative, it can actually lead to a different kind of deception.

Note the mechanical consistency between these two McKay pitches

Brendan McKay has an arsenal consisting of four pitches: A four-seam fastball, a cutter, a changeup, and a curveball. The release points of the four pitches fit within that cluster. This aids the effort of tunneling as the hitter has little to no chance to differentiate the pitches before release.

Whether a pitcher pursues deception by Clippard-like confusion or McKay-like consistency, Release Point Variability makes that effort quantifiable. Ultimately, the goal is to turn indiscernible scatter plots into meaningful metrics. If variability can work in accordance with Release Score and Release Point Dispersion to create insight, it will be fulfilling its purpose.

The complete list of 2019 release point variabilities can be found here:

Leave a Reply

Powered by

Up ↑

%d bloggers like this: