In June, I wrote an article exploring the relationship between the components of a changeup and the amount of groundballs that pitch generates. The model identified vertical movement as the most impactful component, and provided a formula that can be used to identify great changeups, even in a small sample size.
After seeing the results and potential applications of the changeup study, I was interested in applying the same technique to sliders, and attempting to determine which aspects of a slider were most important in getting groundballs.
Although research by Jeff Zimmerman has shown a correlation between slider usage and arm injuries, it can be an effective pitch. It has the highest swinging strike rate (whiffs/pitches) of all pitch types, averaging out at 15.3%, and some pitchers can use it to effectively generate groundballs as well.
Today, I'll focus on identifying the key pitch components in generating groundballs, and projecting groundball rate based on pitch shape. Later in the week, we will look at explaining and projecting whiff rates on sliders.
There are some small flaws with the mechanics of the slider model. Pitchers can throw many variations of their slider, which makes comparing slider-pitchers like Jake Arrieta and Adam Ottavino quite difficult. The PITCHf/x data isn't as clean cut as it is with other pitches.
Ottavino, for example, throws three distinct sliders, something he explained to FanGraphs' David Laurila. One has a "sweeping" movement, one more vertical movement, and one that's "more of a slurve". Even though these are three different pitches, they all get grouped together in PITCHf/x databases as one pitch, potentially causing issues in the data. Regardless, I felt cases like this wouldn't prevent significant conclusions from being drawn, and decided to proceed with the study.
Constructing the Model
To perform this investigation, I first compiled eight years of PITCHf/x data (2007-2014) from Baseball Prospectus' PITCHf/x database. After organizing and formatting the data, I was able to run basic regressions comparing the PITCHf/x components against groundball rates. The PITCHf/x components are velocity, vertical movement, horizontal movement, and "velocity difference" - which is the difference in velocity between the slider and the pitcher's fastest pitch.
After running regressions and removing insignificant variables, I found that velocity, vertical movement and horizontal movement were all significant on the 99.9% level. The r-squared of this regression is .353, which isn't ideal, but it suggests that the model has statistical legitimacy. The following table shows the r-squared values and the p-values for each variable. If you don't care about the math details, feel free to skip right over it.
Vertical Movement | Horizontal Movement | Velocity | |
t-value | -9.28 | -5.41 | -3.28 |
p-value | 6.6 x 10^-18 | 1.44 x 10^-7 | 3.51 x 10^-12 |
r-squared | .088 | .049 | .127 |
Note: All p-values are small enough that we can essentially treat them as being 0.
Velocity was positively correlated, meaning that as slider velocity increases, the theoretical groundball rate increases as well. The relationship is demonstrated in the graph below:
Vertical and horizontal movement, on the other hand, were negatively correlated. This means that as vertical movement increases, the theoretical groundball rate decreases.
It is important to note that vertical movement was significant and negatively correlated in the changeup study as well. Logically, this makes sense, as more drop would cause the hitter to have a higher probability of making contact on the top half of the ball, and driving it into the ground. Further research would be necessary to definitively conclude that more vertical movement is "always good," but so far, it has shown to be beneficial in inducing groundballs on sliders and changeups.
Horizontal movement, however, was confounding. While it was significant, it was negatively correlated, meaning a pitch with little horizontal movement would be most effective at getting groundballs.
When I thought of great sliders, I tended to picture a slider that not only had sharp vertical movement, but sweeping horizontal movement as well. The negative correlation may stem from the angle of contact on the bat and the likelihood of a groundball as a result of it, but this is purely a conjecture. This is something I plan on examining further in the future.
Best and Worst Sliders
After identifying the significant statistics, we can plug data back into the formula to get an expected groundball rate based on its shape. This can be used to identify which pitchers have sliders that are likely to post high or low groundball rates. The formula can be found here, if you'd like to plug in your own values.
Let's look at the best groundball-generating sliders.
Player | Team | Velocity | Horizontal Movement | Vertical Movement | Expected Groundball Rate |
Garrett Richards | ANA | 87.67 | 3.61 | -3.77 | 57.59% |
Brett Anderson | LAN | 82.68 | 2.28 | -5.18 | 56.66% |
Marc Rzepczynski | CLE | 85.61 | 0.82 | -2.25 | 56.06% |
Bryan Morris | MIA | 91.40 | 1.03 | 0.99 | 55.76% |
Jacob Lindgren | NYA | 83.34 | 1.95 | -4.06 | 55.61% |
Blake Treinen | WAS | 87.14 | 2.73 | -2.45 | 55.58% |
Daniel Webb | CHA | 86.49 | 0.57 | -1.10 | 55.09% |
Junichi Tazawa | BOS | 81.94 | 5.28 | -6.84 | 55.09% |
Andrew McKirahan | ATL | 85.14 | 1.07 | -1.91 | 54.53% |
Justin Grimm | CHN | 84.10 | 4.10 | -4.38 | 54.19% |
Now, let's look at the sliders that generate the fewest groundballs.
Player | Team | Velocity | Horizontal Movement | Vertical Movement | Expected Groundball Rate |
Brad Ziegler | ARI | 73.93 | 6.12 | 3.69 | 24.23% |
Joe Smith | ANA | 80.18 | 9.39 | 3.56 | 27.01% |
Sergio Romo | SFN | 78.98 | 9.13 | 2.76 | 27.61% |
Randy Choate | SLN | 75.09 | 5.17 | 3.18 | 27.81% |
Joe Thatcher | HOU | 78.07 | 7.90 | 1.15 | 31.45% |
Rafael Martin | WAS | 81.10 | 6.52 | 3.38 | 32.24% |
Justin Masterson | BOS | 79.19 | 8.95 | 0.39 | 32.78% |
Chris Heston | SFN | 78.05 | 7.26 | 0.76 | 33.06% |
Chris Young | KCA | 80.67 | 3.27 | 4.51 | 33.90% |
Bobby LaFromboise | PIT | 79.53 | 2.07 | 4.01 | 35.25% |
Brad Ziegler's slider looks to be truly terrible, but he doesn't fit the model perfectly because of his submarine arm slot. Over his career, Ziegler has a 36% groundball rate on his slider, which certainly isn't good, but it isn't as poor as it's shape indicates. His submarine delivery is crucial to his success on the pitch.
Angels reliever Joe Smith is another sidearm pitcher whose slider won't be perfectly explained by the model. His groundball rate over his career on his slider is 40%, so his delivery influences his performance on his slider as well. The model isn't as accurate in these extreme circumstances, and is most effective when the pitcher has a common arm angle.
This brings us to Sergio Romo, who throws "over the top" and has an expected groundball rate of 27.61%. He relies heavily on the pitch because of its ability to get whiffs, but it is poor in generating groundballs. Romo demonstrates that while this formula can be useful, groundballs aren't necessary for a pitch to be effective.
Rays Sliders
Now that we've looked at the groundball rates for pitchers across the league, let's look specifically at Rays pitchers. Here is a table of all pitchers in the organization who throw sliders and have PITCHf/x data, and their expected groundball rates on that pitch.
Name | Horizontal Movement | Vertical Movement | Velocity | Expected Groundball Rate |
Chris Archer | 3.12 | 0.50 | 89.06 | 51.33% |
Enny Romero | 0.68 | 0.11 | 85.15 | 51.06% |
Ronald Belisario | 0.59 | 1.37 | 85.52 | 49.10% |
Mark Sappington | 0.15 | 2.49 | 86.12 | 48.13% |
Neil Wagner | 3.99 | 0.96 | 87.04 | 47.02% |
Kirby Yates | 0.54 | 2.69 | 85.54 | 46.57% |
Matt Lollis | 2.02 | 0.16 | 82.02 | 45.70% |
Jim Miller | 2.22 | 1.68 | 83.88 | 44.47% |
Andrew Bellatti | 3.45 | 0.22 | 82.50 | 44.18% |
Alex Colome | 3.49 | 2.11 | 85.43 | 43.63% |
Erasmo Ramirez | 1.16 | 3.76 | 84.88 | 42.89% |
Jose Dominguez | 1.96 | 2.16 | 82.15 | 41.96% |
Steve Geltz | 3.57 | 1.39 | 82.35 | 41.54% |
Ernesto Frieri | 3.89 | 2.10 | 83.10 | 40.53% |
Brandon Gomes | 3.88 | 1.13 | 81.31 | 40.49% |
C.J. Riefenhauser | 6.88 | -1.60 | 78.33 | 38.56% |
Chris Archer has a fantastic slider, but we knew that already. He sits atop the Rays leaderboard, and his xGB% is 38^{th} highest in baseball, out of 395 pitchers. That might not sound excellent, but he is only 2.04% behind the #16 pitcher, and if we look at starters only, his expected groundball rate is 11^{th} highest.
Ronald Belisario's changeup has the highest expected groundball rate in the organization, and his slider grades out as great at getting groundballs as well. This helps explain his high overall groundball rate (60.9% career rate), and his exceptional ability to limit home runs (0.53 career HR/9). Unfortunately, he wasn't able to hold his own in his brief appearance with the Rays, and has elected free agency.
Erasmo Ramirez has been generating groundballs on 59.33% of balls in play, which is much higher than his 42.89% xGB%. I wrote about (and jinxed, I admit) Ramirez and how his increase in groundballs generated from his slider has some legitimacy from his change in pitch location. This graph, created by Jonah Pemstein at FanGraphs, shows that throwing the ball lower in the strike zone causes more groundballs.
Looking at Ramirez's pitch locations, he has been throwing his slider significantly lower in the strike zone this year.
His groundball rate is likely come down as the season continues, but his pitch location tendencies suggest that he may continue to over perform his expected rate.
Despite having many pitchers with sliders that project to generate above average groundballs, the Rays team percentage is 26^{th} in the league, at 43.5%. This is because many of the pitchers with great sliders are in the minor leagues, or have had limited work at the major league level this season. But, it is encouraging to see that the Rays have many young pitchers who have the potential to generate high rates of groundballs, which can help lead them to future success in the majors.
Additionally, this model uses only a piece of the factors of a pitch to project groundball rates. Many other things go into generating groundballs, like pitch location, sequencing, and the hitter at the plate. But, using this formula can give us some idea of the quality of the pitch.
Look for part two shortly, in which we will examine whiff rates on sliders using a similar method.
PITCHf/x data is from Baseball Prospectus and Brooksbaseball.net, and other statistics are from FanGraphs.
Loading comments...