How Biomechanics Affect the Spin Axis of Fastballs
Note from the editor, Chad Raines: This article is written by four of our student analysts. Joey Mylott is a master’s student in biomedical engineering and does biomechanics research in the Wake Forest Pitching Lab. Coby Schneider is a recent graduate from the mathematical business program and will attend graduate school at Northeastern this fall. Naomi Cutter is a recent graduate from the business and enterprise management program and is currently working for the Miami Dolphins in their Business Solutions department. Nihar Maskara is a rising junior in the mathematical business program. All four of these students work on our Wake Forest Baseball Analytics team.
A pitcher’s ability to control the spin axis of their fastball is extremely important to differentiate between specific movement profiles, namely: 4-seam fastballs, 2-seam fastballs, sinkers, and cutters. Manipulating spin with different grips and finger pressures is one way to achieve the desired movement patterns on a fastball. However, the context of how the pitch is released, i.e. the biomechanics of the delivery, also plays a large role in affecting spin axis. For this study, we explored how certain biomechanics metrics affect the spin axis of fastballs for our Wake Forest pitchers.
At first, we just looked at one pitcher with high variability in his fastball’s spin axis and movement profile in order to find the key biomechanics metric(s) that might impact these pitch qualities. Narrowing down the biomechanical movements that most influence his spin axis allowed us to inform the coaching staff of these factors. From there, the coaches had a good idea of specific drills and movement patterns to program in order to guide this pitcher in their desired direction.
After exploring everything just that one pitcher, we expanded this project to include our entire pitching staff to see if we could come to any broader conclusions on the topic. The goal of this project is to create both individual and team-wide markers for our pitchers that can be helpful in creating a more effective and efficient player development process here at Wake Forest.
We will call the individual pitcher we investigated in detail in this study “Pitcher A”. This will keep him anonymous and allow us to refer to him easily throughout the remainder of this piece.
To start, we looked at Pitcher A’s TrackMan data, and a few things jumped out at us as unusual and probably detrimental to his performance.
One of the most concerning findings we found in his TrackMan data was his lack of consistency in certain metrics like release point, extension, spin axis, and pitch movement. In order to see just how variable his metrics were, we compared the standard deviations of these metrics to those of the rest of the pitchers in the NCAA with a similar profile to Pitcher A: a RHP with a release height between 5’ and 6’. To do this, we took all the pitchers that fit this profile and had thrown at least 50 four-seam fastballs in 2021 and found each of their standard deviations for all six metrics of interest. After doing that, we took the median value of every pitcher’s standard deviation so we could see how it compares to Pitcher A. The reason we took the median and not the mean was to control for some pitchers who were outliers and had a very large standard deviation in one of the metrics, which would drastically skew the mean and not be a good representation of the overall group. Therefore, using the median gives us a better understanding of the standard deviation of a comparable pitcher in the NCAA.
What we found was very troubling. Pitcher A has a larger standard deviation than the NCAA median in 5 out of the 6 metrics, which indicates that he is more inconsistent than a typical college pitcher in those specific metrics. If a pitcher was in-tune with his pitching mechanics, you would expect him to have a low standard deviation in all of these metrics since he would have the ability to repeat his motion on every pitch.
After noticing these red flags on TrackMan, we thought that something with Pitcher A’s movements and mechanics might be causing his inconsistency. While there is not a singular correct way to throw a baseball, swing a bat, or perform other sport specific movements, functional movement screens like FMS, OnBaseU, TPI, etc. help coaches identify mobility deficiencies and body awareness issues that cause athletes to excel with some movement patterns and struggle with others. The results from these types of tests can help coaches and trainers maximize athlete performance and player development on an individual level. As expected, Pitcher A’s functional mobility assessment (OnBaseU) indicated some potential issues.
The OnBaseU movement screen results showed that Pitcher A’s main movement issues are poor hip internal rotation on both sides, poor thoracic spine rotational to his follow through side, and a lack of ability to dissociate his hips from his torso in a conscious and controlled fashion. When thinking about some of the key aspects of pitching mechanics, these issues can be problematic. For example, an inability to rotate the hips effectively and efficiently can limit the amount of rotational energy that a pitcher is able to generate and transfer up the kinetic chain. Similarly, limited thoracic spine rotation can lead to less rotational energy from the torso. Combining these potential issues with poor dissociation between the hips and torso can, among other things, seriously impact how much of the energy generated throughout the pitching motion can be efficiently transferred into velocity.
However, the human body is incredibly effective at compensating and making things work within its physical limitations. This is why we saw Pitcher A top out at 98.3 mph with a mean of 95.0 mph on his fastballs this season despite his mobility issues. We believe that these mobility issues may be manifesting themselves in other ways, like his biomechanics and consistency, since velocity and force production do not seem to be a big issue.
We also talked to the Wake Forest pitching coach about his thoughts on Pitcher A’s fastball inconsistencies and mechanics issues. Taking the OnBaseU results and combining them with what he sees from Pitcher A on the mound, he had a few other thoughts on what might be happening. One of the main movement patterns that most amateur pitchers struggle with is properly hinging their hips during the delivery to utilize their legs effectively and efficiently. Pitcher A has a lot of difficulty accomplishing this hinge, which causes noticeable variability in his mechanics later on in his delivery. The pitching coach believed that Pitcher A’s lateral trunk tilt and lateral pelvic tilt are very inconsistent due to his struggle with hip hinging. More importantly to us, he believed that these issues might alter the spin axis of his fastball.
Putting all of these factors and opinions together, we hypothesized that Pitcher A’s mobility limitations in his hips and trunk, especially with regards to core control and rotation, might be the root cause of the wide variability in his fastball. In order to test this theory, we used paired pitches of Pitcher A on our Qualisys motion capture system and TrackMan unit to attempt to connect his biomechanics to his pitch movements.
Two of the main biomechanics metrics that we wanted to investigate were lateral trunk tilt and lateral pelvic tilt. These metrics reflect the positioning of the torso and hips and give a relatively good indication of how the body gets the throwing arm into position during the delivery. As illustrated in Figure 2 below, lateral trunk tilt is the angle of the torso compared to a vertical line. Lateral pelvic tilt is measured in the same way. Also shown is shoulder abduction, which is the angle between the torso and the humerus as one moves their arm away from the middle of their body.
These biomechanics metrics can tell us a great deal about mechanics. A pitcher who has trouble with internal hip rotation and core control probably cannot consistently load into or properly hinge their back hip during their delivery. This could cause them to use their quads and front side more as they load, which could cause imbalances as they get into footstrike. If the pitcher is out of balance and leaning forwards at footstrike (towards 3B for a RHP, towards 1B for a LHP), then a large amount of lateral trunk and/or pelvis tilt could bring the body back to center as pelvic and trunk rotation occur.
Very generally, for a pitcher who keeps their shoulder abduction constant, more lateral trunk tilt at release will lead to a more vertical arm slot and a more back-spun fastball. On the other hand, less lateral trunk tilt at release will lead to more of a side-spun fastball, as shown in Figure 3 below.
We explored a large group of kinematic biomechanical metrics at 3 main events: footstrike, maximum shoulder external rotation, and ball release. The rationale behind using these different time events has to do with the context of the pitching mechanics that we mentioned earlier. The timing and sequence of certain biomechanical positions gives us information about how the pitcher moved from footstrike and continued their delivery through release. Because of how the rest of the body and arm subsequently fall into place, we hypothesized that lateral pelvis and trunk tilt would have a significant effect on the spin axis of fastballs for Pitcher A and the rest of the Wake Forest pitching staff.
For the collection of our data, all evaluations were conducted in the Wake Forest Pitching Lab using a Qualisys marker-based motion capture system. Each pitch was cleaned within the Qualisys Track Manager application and processed using Visual3D. All kinematic and kinetic metrics were calculated within Visual3D using pre-existing pipelines from Qualisys.
The three planes of motion used to describe these metrics are the X, Y, and Z planes. Metrics in the X plane refers to movement or angles in the direction between home plate and the mound (anterior/posterior). Metrics in the Y plane describe movement or angles in the direction perpendicular to the X plane or in the direction between 1st and 3rd base (lateral/medial/side-to-side). Metrics in the Z plane explain rotation about an axis for each particular body segment (rotational).
DISCLAIMER: This project is a retrospective review of Wake Forest pitchers. Unfortunately, pairing the motion capture data with the correct TrackMan pitches proved to be quite difficult. Therefore, our sample sizes were smaller than we would have liked as we only used pitches that we were 100% confident had been properly paired between motion capture and TrackMan. Pitcher A has 25 paired fastballs available to analyze, and the rest of the Wake Forest pitching staff has 44 additional paired fastballs. We only used 9 of Pitcher A’s paired fastballs when looking at the entire team to ensure his data wasn’t weighted too heavily and didn’t skew the results of the larger group of pitchers. We are working to improve our data management system for more complete and efficient pitch pairing between motion capture and pitch tracking technologies in the future.
We initially investigated a number of Pitcher A’s biomechanics metrics and their relationship to his fastball’s spin axis. We created histograms to explore the metric on its own and scatterplots to visualize the relationship between the variables and spin axis. The four variables we will focus on here are:
- Pelvis angle at footstrike (Pelvis_Angle_.Footstrike)
- Pelvis angle at release (Pelvis_Angle_.Release)
- Trunk angle at footstrike (Trunk_Angle_.Footstrike)
- Trunk angle at release (Trunk_Angle_.Release)
For each of these variables, we created graphs of their relationship to spin axis for the X (anterior/posterior), Y (lateral/side-to side), and Z (rotational) planes. In exploring these relationships, we found inconsistencies in almost all of Player A’s metrics that resulted in large ranges of Player A’s spin axis. If Player A’s graphs of his biomechanics metrics vs. spin axis displayed the data points clustered together with a small range in the independent variable, that would indicate that perhaps the metric in question did not have an effect on the inconsistency of his spin axis. What we found was the opposite for most metrics, which suggests that inconsistencies in his biomechanics metrics could be playing a role in the inconsistency of his fastball.
From looking at these graphs, we can see that some of the metrics are pretty stable and others are much more inconsistent. Focusing on footstrike, there is a close clustering of Player A’s pelvis angles in the X and Y directions (forward and side tilt, respectively), disregarding a few outliers. The rest of the pelvis and trunk angles at footstrike are spread across much wider ranges, indicating much higher variability. These results make sense in light of Pitcher A’s problems with controlling his pelvis independently of his torso. We see very similar trends when considering these angles at ball release as well.
The range of spin axis for Player A’s fastballs, which can be seen from the graphs, spans from about 215 to 250 degrees within only 25 pitches. These spin axis values are equivalent to roughly 1:10 and 2:20 when considering a tilt clock. This is quite a large range for a fastball’s spin axis with all pitches thrown with the same 4-seam grip and corroborate the previous TrackMan data from the 2021 season with high standard deviations.
After looking at the general trends in the data, the next step was to create a statistical model in order to find which biomechanics metrics were associated with the spin axis of Pitcher A. Even though we were limited to a dataset of 25 individual fastballs, we sought to uncover some form of a relationship among these metrics. As we collect more paired motion capture and TrackMan data in the future, we will be able to make better and more reliable models for Pitcher A and the rest of the Wake Forest pitchers.
We utilized an elastic net model in order to effectively identify associations with Pitcher A’s data. Elastic net is a regularization technique for performing linear regression that is commonly used when there is risk of correlation and high variance within the data. In this case, a dataset with more than 100 features and less than 30 observations is likely to yield model coefficients with high variability and redundancy among a majority of the predictor variables. By creating a penalty term that restricts the size of variable coefficients, the elastic net technique performs feature selection on the model. This model subsequently identifies important predictors and removes redundant ones, while stabilizing the model coefficients and avoiding high variances. It is commonly classified as a hybrid model of lasso regression and ridge regression, combining the dimensionality reduction powers of the former with the shrinkage techniques of the latter.
The elastic net model for Pitcher A only identified two important features associated with spin axis: Trunk_Angle_Footstrike.1_Y (lateral trunk tilt at footstrike) and Pitching_Hand_Ang_Vel_max (pitching hand maximum angular velocity). However, the pitching hand angular velocity metric is not always accurate because of how the markers are smoothed and filled during the frames around ball release when the hand is moving the fastest. This process often creates inconsistencies with the pitching hand marker, causing the pitching hand kinematics around ball release to be unreliable.
In order to get rid of this metric and avoid overfitting of the data from the elastic net model, we regressed spin axis on Trunk_Angle_Footstrike.1_Y using a simple least-squares regression model. From this model, we found that the individual predictor explains 24.79% of the variability in the spin axis for Pitcher A’s fastballs in our data set. The negative coefficient for Pitcher A’s Trunk_Angle_Footstrike.1_Y shows how his spin axis decreases (gets more backspin) as he gains more lateral trunk tilt at footstrike. This association is an important factor in beginning to understand how the inconsistency of Pitcher A’s biomechanics can lead to a high variability in fastball spin axis. This result also makes sense as we saw a large variability in this metric with a generally negative trend on his spin axis in Figure 4 above. With our limited data set, it seems like our hypothesis about pelvis and trunk lateral tilt impacting Pitcher A’s spin axis was at least partially correct.
These findings for Pitcher A were intriguing, but we thought identifying general trends in these biomechanical metrics among our entire pitching staff could be even more useful. Using the paired motion capture data of fastballs for the entire pitching staff, we utilized the elastic net regression regularization technique again. Only a selected set of variables were considered for feature selection for the full pitching staff: all pelvis angles at footstrike and release, all trunk angles at footstrike and release, extension, release height, and release side. This decision allowed us to isolate these specific metrics that were thought to be related to spin axis while significantly reducing the computing power needed to process the model. In order to correct for RHP vs LHP differences, the metrics with a dependency on handedness (eg. spin axis, release side, and many biomechanics metrics) were reflected so that all pitchers could be properly compared.
After performing elastic net regularization, the final model included each of the variables of interest except the Trunk_Angle_Release_X and Trunk_Angle_Release.1_Y. The fact that Trunk_Angle_Release.1_Y was not found to be significant in this model goes against our original hypothesis, but the Trunk_Angle_Footstrike.1_Y, which was the most impactful metric for Pitcher A’s model, was included, among many other pelvis and trunk angles.
Surprisingly, the elastic net found that these predictors explained 80.83% of the variability in the spin axis of the entire pitching staff. This is a significant improvement from the previous model with Pitcher A, and it suggests that these biomechanics metrics of interest, specifically those of the pelvis and trunk, can have a significant impact on the spin axis for a fastball.
In order to corroborate these findings and determine which features were providing the greatest impact on spin axis, we performed best subset selection, another type of feature selection, on the data. This modeling technique creates n models, where n is the number of features in the data. Starting with a single predictor, the model will find the variable that results in the highest reduction in the residual sum of squares and then use that predictor to create and store a model. This will then be performed for two variables, then three, then four, and so on… iterating until all n features are used and a full model is reached.
The best subset selection model’s effectiveness was notably similar to that of the elastic net technique. It found that a specific group of variables explained 77.83% of the variability in spin axis. The variables and coefficients for the best subset selection model are listed in Figure 6 below.
However, one important detail to note for this best subset selection model is that two predictor variables explained a combined 60.45% of the variability in spin axis. These variables were (1) Pelvis_Angle_Footstrike_X, which is the anterior/posterior tilt of the pelvis, and (2) release side, which is the lateral distance from the center of the mound to the release point of the pitch. Both of these metrics had positive coefficients, meaning that Wake Forest pitchers’ spin axis increased (got more side spin) as their Pelvis_Angle_Footstrike_X became more anteriorly tilted and and their release side got further out to the side. The other four biomechanical variables in the model explained 17.83% of the variability in spin axis.
When comparing the different techniques for Pitcher A and the Wake Forest pitching staff, the models for the whole pitching staff performed much better. A larger sample size of pitchers and pitches could be important factors for this performance increase.
While the models for the entire pitching staff returned a higher adjusted R-squared value than Pitcher A’s model, the results were dominated by Pelvis_Angle_Footstrike_X and release side. These metrics were surprising to us based on our initial thoughts about Pitcher A, suggesting that his results could not be extrapolated to the entire pitching staff.
Release point is very important in determining pitch shape and pitch trajectory, however, it was not initially intended to serve as a primary predictor of spin axis for our purposes. Release point metrics explain the result of the movements that went into delivering a pitch. Biomechanical metrics explain the context of how a pitcher moved through their delivery to release a pitch. In other words, a pitcher can get to the same release point many different ways through different biomechanical sequences. Plus, this metric is not strictly comparable for every pitcher since it can be affected by where they set up on the rubber, arm slot, height, wingspan, etc.
However, the biomechanics metrics still explain a significant portion of the variability in spin axis for Wake Forest pitchers. The most impactful biomechanics metric for this group, according to our model, is the Pelvis_Angle_Footstrike_X (anterior/posterior). This was a surprising result as this movement does not directly impact the body in a way that would affect the arm slot and release side of a pitch. (See Figure 2 above with lateral trunk lilt). Also, interestingly, only one of the pelvis and trunk angles in the lateral/side-to-side (Y) direction at any time point were found to be significant: Pelvis_Angle_Release.1_Y. These results did not support our initial hypothesis.
Keeping our analysis of Pitcher A in mind, increased repeatability of pitching mechanics could make it easier to throw fastballs with consistent metrics. When most of the biomechanics variables of interest remain within a small range over a relatively large sample size of pitches, the spin axis should also only vary slightly. The effects of mechanical changes, when the baseline is so consistent, should become apparent quickly as the new metrics should be distinct from the cluster of previous pitches.
Overall, these results partially confirm our hypotheses that pelvis and trunk angles in the lateral/side-to-side (Y) direction between footstrike and ball release have a significant impact on the spin axis of fastballs for both Pitcher A and the Wake Forest pitching staff. We expected these metrics to be the most influential, but they were overshadowed by other biomechanics metrics and the TrackMan release side metric in some of our models. Essentially, pelvis and trunk angles, in the lateral/side-to-side direction as well as other directions, did have a significant impact on the spin axis of fastballs, but to a lesser degree than we anticipated.
We did find that certain pelvis and trunk angles do play an important role in determining spin axis of fastballs, however, other body segments impact this metric as well. Our analysis did not include biomechanics variables for the distal arm and wrist, which could have a significant impact on the spin axis of pitches. Wrist deviation, forearm pronation/supination, and shoulder abduction are a few movements that might have a large impact on fastball spin axis. Future work could include the distal arm and wrist along with trunk and pelvis metrics for a more in depth and complete investigation. While we don’t have a complete picture of the biomechanics metrics that ultimately influence the spin axis of fastballs, we do have a better idea of some body segments to target when trying to alter it.
