Comparing Release Points in Three Dimensional Spaces

It seems as though almost all media consumed today comes to us by way of ads or sponsorships. Of course, this blog post isn’t sponsored by anything (nor does its traffic warrant such consideration), but if it were to be, it would be sponsored by plotly, a graphics package in R.

While searching for an illuminating way to illustrate pitcher release points in more than two dimensions, plotly presented itself almost immediately. With minimal effort, a random Walker Buehler start could be depicted (as it is below) through the lens of his various release points. The value added by using plotly is the ease of getting that third dimension depicted. The screenshot below captures the vertical (y) and horizontal (x) release points, but additionally offers a pitcher’s extension (z) as well.

This isn’t groundbreaking stuff but, you know, simple pleasures. What it did do was lead to a question regarding how tightly- (or loosely-) clustered pitcher release points are. Given that it brought about this question and thus gave way to the rest of this blog post, one might say what follows is brought to you by plotly, for no reason other than it being a fun and useful tool.

Even in today’s era of lesser innings, it turns out starting pitchers throw a lot of pitches in a season. Given that reality, I used data from Baseball Savant for just 10 pitchers in 2021, the 10 leaders in innings pitched: Zack Wheeler, Walker Buehler, Adam Wainwright, Sandy Alcantara, Robbie Ray, Kevin Gausman, José Berríos, Luis Castillo, Frankie Montas, and Julio Urías. Together, those pitchers threw well over 30,000 pitches in 2021.

With pitch level data from Baseball Savant on those 10 starters, I took a simple measure of center (mean) and measure of dispersion (standard deviation) for all 10 pitchers’ release location data points (horizontal release point, vertical release point, extension), grouped by each pitcher’s pitch classifications.

By examining standard deviations, one can generally glean how dispersed the various release points are by each pitcher’s pitch type. One can see who has the most repeatability on their individual pitches and who potentially alters their arm slot (intentionally or otherwise) in games. All data points are presented in inches.

Below is a look at those pitches that were thrown at the most consistent release points, as measured by the average standard deviation across all three release point measures. Taking the average of three standard deviations isn’t exactly rigorous, but it does the job for this general exercise. This Top 15 starter-pitches chart presents, you guessed it, 15 of the most consistently released pitches among the wider 48 starter-pitches in this dataset.

Robbie Ray’s sinker in 2021 was released at incredibly consistent points: the standard deviation in all x, y, and z directions was just roughly 1 inch. Meanwhile, all 6 of Adam Wainwright’s pitches seem to have exceptionally regular release points (between 0.8 – 1.8 inches in any direction); clearly, Wainwright wasn’t one to alter his arm slot in order to offer batters variable looks in 2021.

Other points of interest arise from the average release data. For one, this chart illustrates the elite extension Zack Wheeler gets on his pitches, specifically here his sinker which has ~6-10 inches more extension than the other pitches listed. Additionally, one can see the relative horizontal release of a pitcher like Walker Buehler (more over the top) versus José Berríos (lower, more horizontal, arm slot).

Next, below are those pitches with the greatest average standard deviation across horizontal, vertical, and extension release point data.

Julio Urías’ sinker can be disregarded from this leaderboard as Baseball Savant classified just 18 of his pitches in 2021 as sinkers, which leads to sample size complications. Across these 48 distinct pitches by 10 pitchers, just less than an inch separates the most variable (2.08) from the least variable (1.12) average standard deviations, indicating again the incredible consistency across all these pitchers. Without further context, it is tough to really qualify these measures, but as a lay person, standard deviations of release points being just ~1-2 inches in any three-dimensional direction is pretty incredible. Repeatability of motion is very clearly an integral part of pitching (shocking, indeed).

Those two charts above in part covered variations within starter pitch types, but what about variation across starter pitch types? In order to very gently scratch the surface of that topic, I have calculated the euclidean distance (i.e. “as the crow flies”) across all permutations of pitch types for three pitchers: Alcantara, Berríos, and Buehler. In other words, taking the average release point (on x, y, z axis) of each pitch per player, what is the distance between average release point for pitch X and the average point for pitch Y, and for each permutation of offerings within each pitcher’s arsenal thereafter?

The juxtapositions, presented in distance, between pitches presents an interesting view. Sandy Alcantara releases all his pitches (or rather, the average location of those release points) in an incredibly tight cluster. Without concrete data to follow up on, it is easy to speculate that such similar release points, on average, across pitches has to mean positive things for Alcantara’s deception in mixing his pitches. Berríos, meanwhile, appears to be in the middle of the pack, of this group of 3 at least; there is slight variation – from 0.5 to 3 inches – in his release points across pitches.

Here at least, Buehler appears to be the outlier. His release points across pitches are quite varied relative to these other two starters. His knuckle curve in particular deviates in release point from several other pitches, namely his changeup, four-seam, and sinker. If you are to reexamine the first 3D scatterplot above of Buehler’s release points for a single start, it appears that, on that day at least, his curveball was released with greater extension and lesser vertical height relative to his other pitches.

In order to wrap things up and offer a bit more context, the visualization below graphs the average release point by pitcher, across all 10 pitchers mentioned in this blog post.

One only needs to glance at the distribution of Walker Buehler’s reddish-brown points, each of which represent the average location of a specific pitch type, to see they are more widely distributed than most of the other player-pitches here. In particular, Frankie Montas, Zack Wheeler, Sandy Alcantara, and Kevin Gausman are very tight distributions of their release points across pitches relative to the other pitches depicted.

None of this is to suggest lesser variety in release points is strictly better; 10 pitchers is a sorry sample size to support conjecture anyway. Additionally, Walker Buehler was a Cy Young contender in the NL this season. In fact, all of these pitchers had considerable success in order to hit the innings thresholds that they did coming off a shortened 2020 season. In all, this has been about discussing release points in a way I have not seen elsewhere to any great extent. With any luck, it will beget additional lines of questioning.