# Investigating Cade Smith’s Splitter

About two weeks ago, I posted a thread on Twitter showcasing the five most likely pitches to result in a swing and miss based on my models from 7/23 to 8/2.

It is worth noting that all five pitches were splitters. This should not come as a surprise, considering splitters have the highest swing and miss rate among the eight pitch types (my models account for).

2015-22 (training data)

2024

x_swing_and_miss_percent is expected swing and miss percent. For a given pitch, expected swing and miss probability is calculated by multiplying an expected swing probability by the expected probability that the swing would result in a whiff. The two factors/probabilities are outputs from two logistic regression models. The first considers all pitches of a particular pitch type, the second only swings against that pitch type. In total, there are 16 models.

- CH (changeup) swing model
- CH whiff
- CU (curveball) swing
- CU whiff
- FC (cutter) swing
- FC whiff
- FF (four-seam fastball) swing
- FF whiff
- FS (splitter) swing
- FS whiff
- SI (sinker) swing
- SI whiff
- SL (slider) swing
- SL whiff
- ST (sweeper) swing
- ST whiff

The inputs to the models include pitch characteristics like these, or variations of them.

- p_throws (pitcher hand)
- release_pos_x_adjusted (horizontal release position adjusted for batter hand)
- release_pos_z (vertical release position)
- sqrt(release_pos_x^2+release_pos_z^2) (hypotenuse of release_pos_x and release_pos_z)
- release extension
- release_speed
- release_spin_rate
- release_spin_rate/release_speed (Bauer units)
- pfx_x_adjusted (horizontal movement adjusted)
- abs(pfx_x)
- pfx_z (vertical movement)
- abs(pfx_z)
- sqrt(pfx_x^2+pfx_z^2)
- vx0_adjusted (horizontal velocity adjusted)
- abs(vx0)
- vy0 (velocity in y-direction)
- vz0 (velocity in z-direction)
- abs(vz0)
- sqrt(vx0^2+vy0^2+vz0^2)
- ax_adjusted (horizontal acceleration adjusted)
- abs(ax)
- ay (acceleration in y-direction)
- az (acceleration in z-direction)
- sqrt(ax^2+ay^2+az^2)
- plate_x_adjusted (horizontal position of pitch at plate adjusted)
- abs(plate_x)
- plate_z (vertical position of pitch at plate)
- abs(plate_z)
- sqrt(plate_x^2+(plate_z-2.5)^2) (distance from center of plate)
- is_in_zone (pitch is in [strike] zone)

Basically the same list can be found in An update to expected swing and miss%+, an article I wrote last season. The only real changes I made earlier this year were expanding the training dataset to include the 2022 season and running the models on sweepers.

Returning back to the five splitters from beginning, two of them were thrown by Cleveland Guardians reliever Cade Smith.

Of the 53 pitchers to throw at least 100 splitters in 2024, Cade Smith ranks first with a 26.3 expected swing and miss percent. In reality, he is only getting swings and misses on around 13 percent of his splitters, not even half of the expected rate. Interestingly, only Nick Sandlin, Smith’s teammate, has a bigger difference between his actual and expected percentages (28.1 vs 13.2), and his is in the opposite direction.

The question I want to try to answer is this one.

In other words, why does my model love Smith’s splitter and why is he not getting more swings and misses with it?

It is not because he is right-handed. Here are splits from 2015 to 2022 for splitters….

What I want to do is compare Cade Smith’s 2024 splitters to the top and bottom 10 percent of splitters from 2015-22 in terms of expected swing and miss percent.

A little bit of a lower release point.

He does noticeably get more extension.

Not sure if it is for the better, but his splitter does seem harder than most.

Less spin.

Less spin per mph.

Appears as though some of the splitters in the bottom 10 percent are dropping too much. Smith’s density plot lines up well with the top 10 percent one.

Not too much movement.

Greater horizontal velocity than the top and bottom groups.

Moving faster vertically too.

Traveling faster overall.

Accelerating faster vertically than the top and bottom 10 percent.

Greater overall acceleration.

Not locating his splitters too far in or out, for the most part.

The bottom 10 percent of splitters are located too high up in the zone or too low, while 8,051 of the 8,052 in the top 10 percent are in the bottom half of the zone, the majority in the vicinity of the bottom of the zone. Cade Smith seems to locate his splitter well.

His splitters are not too far from the center of the plate, which theoretically translates to more swings.

Some additional (uncovered) measures to examine for the whiff model…

The lower vertical release point can be seen again.

Faster.

Less spin. Same themes as before, so I am not going to go through the rest of the polynomial terms.

To summarize Cade Smith’s splitter, it drops an above average amount but does not break much to the side, comparatively speaking.

His splitter is also harder than most (13th fastest out of 53 on Baseball Savant).

If I had to provide one reason as to why the model loves Smith’s splitter, it would be its location.

On the graph below, it is clear that Smith’s splitter this season is an outlier (the other points are splitters from 2015 to 2014, minimum 100 thrown). I anticipate that he will get more swings and misses on the pitch moving forward.

Possibly one reason for the discrepancy in his actual and expected swing and miss rates on the splitter is that he throws his fastball nearly 70 percent of the time, so hitters are not looking for the splitter. Or maybe it does not play well off of the fastball. It also could be that the model is wrong!

Featured image from MLB.com