The Rays Way: Varying individual pitch speeds

Last season, the Rays implemented a pitching philosophy that garnered national attention. A large portion of the pitchers on the roster had high vertical movement measurements, or "rise" on their fastball, and the team instructed the pitchers to throw their fastballs high in the strike zone. The rise on the fastballs complements the high location in the zone, as hitters swing underneath the pitch, and whiff or pop up.

Previously: The Rays Way: Rising Fastballs

This was a deviation from the trend around the league of attacking the bottom of the strike zone, which was an effort to prevent home runs and instead generate ground balls.

By the end of the season, the Rays had clearly implemented this philosophy throughout their pitching staff. As a team, 20.91% of their pitches were high fastballs, which was the highest rate in the league. Individually, more than half of Rays pitchers had high fastball rates above league average, and those that were below league average were only a few pitches away.

This approach helped the Rays maximize the performance of each pitcher, and is a clear example of the organization diverting from the trends and "fads" across the league.

In general, the Rays certainly aren't opposed to making changes in approach that didn't necessarily follow what other teams were doing. In this year's edition of the Baseball Prospectus annual, Chris Mosch lays out numerous examples of the Rays finding underutilized and unique philosophies to help the team, like drastic outfield shifts and alignments, pulling starters after they go through the order twice, and pitching up in the strike zone. During the 2015 season, the Rays introduced another yet another wrinkle into their pitching philosophy.

Pitchers like Chris Archer and Nathan Karns were throwing one of their pitches at different speeds. Archer effectively turned his slider into two sliders, and Karns had two changeups with distinctly different velocities. However, this change wasn't being picked up on Pitch f/x cameras, because the movements were very similar and there was significant overlap in the speeds of the pitches.

Using raw Pitch f/x data from Baseballheatmaps.com, I was able to create visualizations to illustrate the differences in speeds for these pitches using Kernel Density Estimation.

Kernel Density Estimation

Similar to a histogram, Kernel Density Estimation (KDE) is a method of displaying data. To show evidence of Archer and Karns varying speeds of their slider and changeup, respectively, we would need to see evidence that the data is bimodal. This could be done using a histogram, but histograms have some drawbacks.

When using a histogram, the bins of the data work in absolutes - a pitch is either in a bin, or out of it. Because of this, it can mask the true distribution of the data. For example, here are two histograms of Archer's slider velocity, using the same set of data:

Even though these histograms are created using the same set of data, the distributions look different. In the first histogram, the data looks bimodal. In the second histogram, however, the data looks unimodal. Because of these inconsistencies, histograms can lead to ambiguous answers.

With KDE, instead of dividing the data based on arbitrary bin widths, a "kernel" of normal distribution is placed at each point of data. The kernels are then added together, and the KDE is formed. The process is demonstrated in the image from Wikipedia below. The red dashed lines represent the kernels, and the solid blue line on the right represents the KDE.

Using KDE generates a smoother image, and allows for a more accurate representation of the data.

Chris Archer

With this KDE for Chris Archer's slider, we see two clear "peaks" in the data, suggesting there are elements of bimodality, and that Archer varies the speed of his slider. Essentially, we have this (Image adopted from Hardball Times):

Judging Archer's slider based on its shape, it already grades out to be one of the best in the league in both whiffs and changeups. Changing the speed of his slider adds another element to the already dominant pitch, and helps explain how it was so effective last season.

During his time in the minor leagues, analysts were skeptical if Archer's arsenal was wide enough for him to remain as a starter. His slider and fastball were labeled as above average to plus, but his mediocre changeup cast doubt on his ability to get through the order multiple times. Since then, Archer has improved his changeup, and turned it into a serviceable pitch, but he also expanded his arsenal by varying the velocity of his slider.

In changing speeds, Archer effectively turns his slider into two pitches. While it doesn't have the platoon advantage that his changeup has, it allows him to keep hitters off balance and improve the results he sees on the pitch.

Furthermore, Archer didn't vary the speed of his slider last season. Looking at his KDE for 2014, there is a clear unimodal distribution to the data.

While there are certainly many factors that went into Archer's further breakout last season, and correlation doesn't always equal causation, changing speeds likely contributed to his excellent season.

Nathan Karns

Karns, like Archer, was initially labeled as a candidate to be moved to the bullpen because of the lack of a third offering - scouts liked his fastball and curveball, but questioned whether his changeup and command would hold up in the rotation.

Fortunately, it would seem, Karns made significant improvements to his changeup last season. He added more drop to the pitch (less vertical movement), and saw increases in both whiff and groundball rate. But the component that allowed him to turn his changeup from average to effective was changing the speed of it.

Karns' KDE shows that he has thrown two distinctly different changeups. From the bimodal distribution we can see he is able to create a velocity difference of nearly two mph on the changeups, making it more effective.

Furthermore, slow and fast changeups are better at generating different outcomes. Velocity is positively correlated with groundball rate, meaning faster changeups get more groundballs. However, whiffs are positively correlated with the velocity difference between a pitcher's fastest pitch and their changeup. Holding the fastball velocity equal, a slower changeup would have a greater velocity difference, leading to a higher expected whiff rate.

In having a changeup with two different speeds, Karns is effectively maximizing the potential outcomes of his changeup. The faster changeup will generate groundballs at a higher rate, and the slower changeup has a higher expected whiff rate. There is likely a performance benefit from the interplay of the two pitches as well.

Of course, the Rays won't be able to reap the benefits of Karns' changeup development moving forward after his trade to Seattle for SS Brad Miller, but he does serve as a good example of how the philosophy of varying speeds can be implemented effectively, and how it can improve a pitcher.

Conclusion

Overall, by instructing a pitcher to throw one of their pitches at multiple speeds, the Rays can increase the size of a pitcher's arsenal without introducing what is effectively a new pitch.

Both Archer and Karns saw significant increases in performance after starting to vary their speeds for an individual pitch. While other aspects were apart of their improvement, changing pitch speeds certainly contributed to their success.

The philosophy has likely been extended to other Rays pitchers as well.

Jake Odorizzi, for example, may be doing something similar with his fastball, but because he throws two pitches between 88-92 mph, it is challenging to differentiate between an intentionally slower fastball and a cutter without individually evaluating 1500+ pitches.

More research is certainly needed on the topic. It is unclear if all Rays pitchers are doing this to some extent, or just those that would benefit the most from an expanded arsenal. Regardless, Archer and Karns serve as evidence that this approach not only can be implemented but greatly benefit the pitcher.

Pitch f/x data is from Baseballheatmaps.com. Statistics are from FanGraphs and Baseball Savant.