Finding Success with Changeups
During the shortened 2020 season, Milwaukee Brewers rookie reliever Devin Williams dazzled spectators with an incredible pitch that would later be coined the name “The Airbender”. If you are not familiar with this new phenomenon brought to our attention by the 26 year old, here is what you have missed:
Already one of the most dominant pitches in the game today, The Airbender was initially classified as Williams’ changeup. However after making each hitter look more foolish than the last, Williams began to label it in a league of its own, despite tracking data still having labeled it as a changeup. What makes this incredible pitch so difficult to understand and even more difficult to make contact with? Surely it has to be the obvious movement and late break the ball takes during its path from Williams’ hand to the plate.
While the rest of the league’s pitchers attempt to replicate The Airbender in their own fashion, pitchers still need to be able to find ways to throw effective changeups without the ability to defy physics as The Airbender does so casually.
But how exactly does a pitcher create a changeup that performs so well, and what does this formula look like? It is not possible to create one specific profile of a changeup that will work for every pitcher, as traditional changeups (the ones that don’t “bend air”) are built on the foundation of disrupting hitters’ timing and creating deception. Both of these ideas are made possible depending on a pitcher’s fastball. This post will look into what makes an effective changeup in comparison to different fastball profiles.
Background
The first thing I had to determine was how to properly define what I wanted this “effective” changeup to look like. How fast should it be thrown? What kind of movement does it display? What does it pair well with? How often is it thrown? The short answer: all of the above.
Keep in mind that the purpose of this project was not to discover a particular profile for what every pitcher’s changeup should look like, but rather to identify what features of a changeup are most effective, based on the profile of a pitcher’s fastball. So while we want to look at the pitch’s various features, all of it is subject to change depending on what the fastball looks like, and even what type of pitcher it applies to (more on this later).
Metrics included were those that were deemed necessary and important for the purpose of obtaining significant results, but also for convenience of what was provided in the given dataset. While there are many more things we can look at, the root of it consists of the following features:
- Velocity (MPH)
- Spin Rate (RPM)
- Bauer Units
- Horizontal Break (ft.)
- Induced Vertical Break (ft.)
- Release Side (ft.)
- Release Height (ft.)
- Ratio FB/CH
And what exactly are we looking for to prove effectiveness? Whiffs, ground balls induced, xBACON and xwOBACON were the four I chose to settle on for now. All four metrics were later examined as averages per changeup during the analysis.
As far as the data itself goes, I looked at MLB Statcast data from the 2019 season. Due to the small sample size and irregular conditions encountered during the shortened 2020 season, this data was excluded and the decision to primarily focus on the 2019 season was reached.
Something to note as well before diving too deep is that four seam and two seam fastballs were grouped together as one classification of fastballs.
Methodology
Given the objective of this project, I decided that the best way to observe the features of pitches was to look at the separation gaps for each feature between a fastball and changeup. Doing so would allow for this “comparison analysis” aspect when observing each pitcher’s changeup in comparison to their fastballs.
I then chose to filter pitchers to only include those with larger sample sizes for both fastballs and changeups during the 2019 season. Within the MLB Statcast dataset, pitchers observed were those with at least 500 fastballs and 100 changeups thrown.
The comparison analysis began by binning each of the variables, with the intent to observe potential patterns in the outcome variables. The sizes and ranges of each bin varied by each metric, as the end goal was to best balance out the number of samples in each. Outcome variables of whiffs, ground balls, and expected contact metrics were observed in each of the tables as they were created, with the goal of identifying any potential trends or patterns in the data. For example, looking to identify how the outcome variables were affected with an increasing or decreasing gap in velocity.
To further prove if some of these trends held true when applied in a different context, I took a look at those that performed well in each of the outcome metrics, and those that performed not-so-well. I created tables for the top 10 and bottom 10 pitchers in the MLB for each of the outcome metrics, comparing observations and patterns found in those data frames with the data from the binned tables. Most of my observations came from identifying if a pitcher deviated from the averages of each variable that pitchers typically fall in.
Analysis
Going into this project, I had learned of quite a few theories about how these variables affect changeups, and which ones provide the best production.
- Velocity gap between fastballs and changeups has a significant effect on changeup performance
- Horizontal Break is more important for changeups, with induced vertical break having little correlation to success
- Having similar release points for both fastballs and changeups is advantageous for creating deception
Keeping these in mind during the research helped me to understand more of what I might be looking for with each of these. We’ll start with velocity gap:
Velocity gap
With no clear expectation of how velocity gap would play a role in performance going into this project, it became pretty clear there would be a solid relationship established with whiffs.
Looking at the table above, as the velocity gap between fastball and changeup increases, the average number of whiffs per changeup increase as well. On the flip side, with this increase in velocity gap, also seems to be a decrease in rate of ground balls.
A larger velocity gap seems to benefit those with the best changeup whiff rates in the MLB:
All pitchers on this list have velocity gaps in the average range of 6–8 mph at the minimum, with most of them at the higher end of that range. Five of the ten even have velocity gaps larger than 8 mph, with a couple reaching up to a 12 mph difference. It’s also noticeable that seven out of the top ten have fastballs 95 MPH or more, an ideal way to induce whiffs.
However when taking a look at the other end of the spectrum, with the bottom 10, we see an almost complete opposite version of the results:
Only two of the ten have fastball velocities at or above 94 mph, and all lie toward the lower end of the 6–8 mph velocity gap range, with a few even falling slightly below.
However, the narrower velocity gap seems to benefit those with the best changeup ground ball rates:
Notice that all but two of the pitchers in this list have fastballs averaging 94 mph or greater, and nearly all of the pitchers in this list lie within the 6–8 mph average velocity gap range, with many of them sitting on the “lower” end of the spectrum around 6–7 mph, and a couple even falling below at 4 and 5 mph.
Mychal Givens is an interesting case, standing alone as one of the pitchers in this list with a velocity gap higher than the average, and the only pitcher that appears on both top 10 lists. The numbers tell a story, but seeing that changeup paints a picture:
We see that larger velocity gaps are not always harmful to ground ball production, such as in the case of Mychal Givens, and there still are ways to be productive with these larger gaps.
The velocity gap variable seems to trend as expected for the bottom 10 ground ball pitchers, with a lot of larger velocity gap differences and many well above the average:
Additionally, velocity gap seems to be an influential factor for the pitchers with the best expected contact numbers as well:
Overall, it seems as though the pitchers in both of these expected contact lists fit the profile of those from the successful whiff pitchers. High fastball velocity and larger velocity gaps seem to be a common trend on these lists as well.
Movement Differential
Moving on to movement differential, ideally we would like to see a pattern somewhere in the horizontal break data. However what we get is not exactly as promising as we might have thought or prior hypotheses might have previewed:
Here, we would expect to see the highest number of whiffs and ground balls induced at the top of the table, where the gap indicates between 0.6–1.5 feet of horizontal break separation, and the lowest numbers towards the bottom where there is the smallest gap in separation. Instead, we get results that do not indicate that horizontal break has much of an effect on the outcomes, as there is no clear trend upward or downward for whiffs induced or ground balls induced as the gap in horizontal break widens.
An additional part of the previous hypothesis was that induced vertical break has little correlation with success and that mis-hits by the batter are more likely to get hit harder when the pitch has more induced vertical break.
This becomes interesting, especially when taking the hypothesis into account, as we see a potential positive relationship between induced vertical break gap and ground ball rate:
As the gap of induced vertical break gets larger, we see a consistently increasing percentage of whiffs and ground balls induced per changeup. And we see the same apply when looking at the top 10 ground ball pitchers:
With the average induced vertical break gap of all pitchers between 0.3–0.6 feet, seven out of the ten pitchers on this list have greater than 0.6 feet of separation between their fastballs and changeups, with four of those having more than 1 foot of separation.
Additionally, the bottom 10 pitchers seem to lack a large enough gap in induced vertical break unlike the pitchers of the top 10 list. All but one of the 10 on this list land somewhere in the average range of induced vertical break gap, or somewhere below it.
Release Point
Moving on to the analyzing pitcher release points.
The binned data suggests that a larger gap in Release Height would allow pitchers to benefit from a ground ball production standpoint, however slight fluctuations still remain and make it hard to sell:
What is perhaps the most interesting observation when considering Release Points, is the gaps for both of the bottom 10 expected contact lists:
All pitchers on these lists seem to “mirror” their release points of both pitches by narrowing the gap in Release Height and Side, which given that these are the pitchers who performed poorly in this aspect, seems as though we could be going further against one of the initial hypotheses.
Now looking at both of the expected contact top 10 lists:
Highlighted here are the gaps that deviate further from 0. It’s very clear that by comparing those in the top 10 list versus those in the bottom 10 list that the bottom 10 list mirrors their release points much better than those in the top 10. However, this “better” release mirroring does not seem to perform very well when it comes to expected contact, telling us that a slight deviation in release point leads to improved xBACON and xwOBACON per changeup numbers.
While the data might suggest a pattern here, this is difficult to draw any conclusions from. Further research would be necessary to help explain this trend.
Conclusions
To wrap everything up, I want to look at what can be taken away from this analysis. We can divide our conclusions into two categories, inducing more whiffs, and inducing more ground balls.
Whiff:
- Higher fastball velocity guys will want to have a larger velocity gap between their fastball and changeups.
Groundball:
- Lower fastball velocity will look to have smaller velocity gap between their fastball and changeups.
- Creating a larger gap in induced vertical break between fastballs and changeups contributes to ground ball success.
Applications
Taking all of this research provided, what can pitching coaches and players take away from this changeup comparison analysis?
I believe all of this can be factored in when talking about changeup pitch design. However, as reiterated throughout this post, it will differ from one pitcher to another.
There are a lot of ways pitchers can find success with their changeup, in whatever way they choose to define this success. While I specified my research to two main outcome variables that I used to define “success”, whiffs and ground balls, I am sure there are many other ways that pitchers can achieve success with their changeups in other ways not explored in this post.
Some pitchers can find multiple ways of success due to their unique combination features that makes their pitch great (i.e. Mychal Givens).
While it seems rare to find a pitcher that can do it all, Givens’s above average velocity gap, paired with a solid 95 MPH fastball, and above average Vertical Movement differential seems to play very well for him and his below average arm slot delivery. Givens was also a great example of having varying Release Heights between his fastballs and changeups, making his combination even more unique, and quite successful.
For the sake of this research analysis, I am going to focus the applications to two groups of pitchers: the high-velocity and whiff-inducing types, versus the low-velocity and ground ball-inducing types.
A pitcher that has a high fastball velocity and is looking to make the most of their changeup could look into slowing down their changeup, thus creating a larger velocity gap between the two, in order to produce whiffs.
Or, a pitcher that is determined to increase their whiffs might look to adjust their changeups accordingly, despite whether or not they have initial high fastball velocities.
In contrast, a pitcher that is on the lower end of the spectrum in terms of fastball velocity might look to create a firm changeup by decreasing the gap in velocity, while increasing the gap in induced vertical break. These guys might struggle with creating as much whiffs due to the lack of fastball velocity (i.e Dallas Keuchel or Daniel Norris), but can still make their changeups more effective.
A ground-ball type of pitcher might look to apply these concepts regardless of their fastball profile as well.
Limitations
Throughout the course of this project, I ran into several limitations that should be taken into account.
While not necessarily a limitation, a reminder that this research is not a blueprint for developing a changeup. There is no one correct formula for a pitcher to have success, especially when it comes to how one crafts their changeup. It could vary greatly from pitcher to pitcher, and we saw this stand out when comparing the whiff-heavy pitchers to ground ball-inducing pitchers, where those with higher velocities were more likely to be the whiff type and those with lower velocities tended to be the ground ball types.
Next, binning and sorting out the lists of pitchers helped identify possible trends and patterns that existed with the dataset, yet there were plenty of outlying instances that defied these trends and could possibly be looked at with further analysis in future research. We saw this with two of the top 10 ground ball pitchers, Jacob Waguespack and Dallas Keuchel, who despite a far narrower induced vertical break gap between their fastballs and changeups, still made the top 10.
Next Steps
In this post, I explored a lot of ways pitchers find success with their changeups by comparing their changeup profile to that of their fastball, which allowed up to better understand a sort of cause and effect relationship between the variables and how I chose to define success. Yet, there are still many things I did not get around to researching for this post, but would be worth looking into in future research to get a better, more in-depth changeup comparison analysis.
These include:
- Further analyze changeup effectiveness by arm slot type (over the top, 3/4, sidearm, submarine, etc.).
- How might arm speed play a role in improving deception.
- What do changeups play best off of in terms of sequencing, and does this differ for HB vs induced VB-heavy changeups.
- How does Vertical Approach Angle affect these results, and can it help determine where changeups can most effectively be located.
- And of course, how Seam Shifted Wake might play a role.
Lots of interesting things to look into, hopefully this post can soon be renamed to “Finding Success with Changeups Pt. 1”.
