How to Predict Launch Angle, and What it Can Tell Us
It’s still winter break, and as I’ve been preparing for the spring slate of games at Arizona, I often get ideas for projects that sound interesting… This is one of those projects.
It is often theorized and relatively proven that the launch angle of the ball is controlled more by the pitcher, than by the hitter. Assuming that is true, we should be able to predict the launch angle of a particular pitch by the pitch characteristics. So let's do that.
No code this time, sorry… If you have direct questions about my methods/code feel free to DM me on Twitter/Instagram and I’ll try to get back to you!
For the model, I used data from 2019–2021 for no reason other than convenience, and honestly, that is likely a large enough sample size. I ran an XGBoost for three different pitch groupings using the following characteristics that provided the importance matrix below:
Fastballs:
Four-Seam, Two-Seam, Sinkers
Offspeed:
Changeups, Splitters, Eephuses
Breaking:
Sliders, Curveballs, Cutters, Knuckle-Curves
You’ll notice each model contains a different number of variables, and I did that so each variable would remain significant (> 5%)
The first observation that’s important is which variables even matter for each grouping and how do they differ.
For Fastballs, it’s pretty much a two-variable show with vertical break and plate height. Plate side and horizontal break are likely there for some sinkers. But this makes sense, along with release height, also being significant, it comprises the main variables in Vertical Approach Angle. A metric that explains why many fastballs are/are not successful in certain parts of the zone, launch angle being no different.
Offspeed has the most variables by a decent margin. While plate height and vertical break lead the pack, the horizontal plate location is also very important. Changeups make their money on moving from side to side, and the location of such can produce groundballs, and thus low launch angles. Vertical Movement Difference from that pitcher’s fastball (vmov_diff) is also important. The more separation a pitcher can make with their vertical movement between pitches can allow them to control the launch angles they allow.
With breaking pitches it’s really only one variable that is important, plate height. If you leave a breaking pitch up, it’s likely to be hit in the air… The rest is pretty expected, similar to offspeed.
Now let’s take a closer look at different pitches and their variables:
Four Seam Fastballs
For reference: the dotted lines are the top and bottom of the strike zone, and the red line is the average vertical break (or whatever variable is on the x-axis)
With Four-Seam Fastballs the trend is quite clear, more VB and higher on the plate equals a higher expected launch angle. That was obvious before, however, what is interesting is what’s in the middle. We can see that with plus vertical break, plate height becomes less and less of a factor, indicating that if you have that plus jump, it’ll play everywhere.
Sinkers
Two plots for sinkers here, one for the horizontal break, one for vertical. We see a similar pattern to fastballs. If a pitcher has plus sink or plus run, they can maneuver with more space around the plate and still get those ground balls that sinkers are often looking for.
Changeups
Changeups look quite similar to sinkers but appear to have a little less leeway when it comes to missing up/down, but the principle still holds overall I’d say as plus movement creates more launch angles close to the extrema (> 30 or < 0).
Sliders
With sliders, we see a distinct juxtaposition between offspeed and fastballs. As seen in the importance matrix for breaking balls, plate height is the main determinant of launch angle. While you can produce high LAs low in the zone and low LAs up in the zone with plus movement, you really want to keep those sliders down unless they have significant drop or sweep.
Curveballs
Curveballs are interesting as plate height doesn’t appear to be as much of a factor overall, at least relative to sliders. Movement plays a large role here, especially with positive vertical break curves (gross). Also interesting is the relationship with horizontal break, a place that appears where a pitcher can heavily control the launch angle (sweepers ftw).
One more thing before I go, let’s look at individual players and pitches. I posted this on Twitter, but here are the highest and lowest xLA pitches since 2019:
For individual players, we can compare their xLA to their real-life LA in 2021 to see if regression is due, and before that, it only makes sense to provide the overall correlation plot:
Some outliers at the ends, but I’m pretty happy with how this turned out overall, a pretty strong, linear correlation here with an R² of 0.774.
Biggest “Over-Performers”
Sorry about the accents in the name
These are the players whose xLA is much higher than theirs in real life… Maybe expect some more line drives and fly balls next season.
Obviously, to predict the scale of such regression, further analysis on a micro-level is needed.
Biggest “Under-Performers”
Gonna be honest, never heard of Steven Okert, but he’s on the Giants fun fact
Lots of relievers here compared to the number of starters in the other list, bt Crichton probably stands out the most here with an increase in ground balls.
Highest Average xLA
Lowest Average xLA
I’d say that lower xLAs are more interesting to look at because many of those dudes make their money on ground balls, while many higher launch angle guys rack up the Ks and give up more fly balls.
Framber Valdez will still probably be a super ground ball merchant, but maybe a little less next season? Who knows?
Highest xLA By Pitch:
No surprise here, all fastballs with plus VB and/or a lower release slot.
Lowest xLA By Pitch:
Also not much surprise here, sinkers and offspeed designed to force ground balls. Joe Smith with his absurd -16.7 LA clearly is not sustainable, but he still ranks top in xLA.
For future work on this, one could use this number to directly predict ground balls, line drives, fly balls, etc… That’s likely the next step here to evaluate changes on a scale that is more directly correlated to outs and run value. Maybe in the future!
Well, that’s all I have for you today, if you have any questions, comments, feel free to tweet at me, DM me or just contact me in general!
Follow me on Twitter: @JacksonDels2
On Instagram: @jacksondels
