Identifying great changeups

It was a Tuesday night in Atlanta, and the Braves were hosting the Rays. The Rays got off to a fast start, as Kevin Kiermaier roped a triple to start the game, and then scored when Steven Souza Jr singled in the following at bat. But, Erasmo Ramirez had trouble gripping his fastball, and after the Braves scored three times in the second inning, the Rays found themselves in a 3-1 hole. The Rays chipped away at the lead, and eventually pulled ahead when David DeJesus doubled, driving in Logan Forsythe. Entering the top of the ninth, Tampa Bay was up 5-3.

Brad Boxberger started the inning and set the first two batters down quickly. Adonis Garcia stepped up to the plate. After a few pitches, the count moved to 2-2. Garcia needed to get a hit to keep the game alive. Maybe Garcia was looking for a fourseam fastball, because Boxberger likes to throw those when ahead in the count. Boxberger deals, and . . .

(Video courtesy of MLB.com's Youtube channel.)

BAM. Changeup.

There are many benefits to throwing a changeup. They can be used to effectively generate groundballs and whiffs, as Boxberger shows at the end of that game. Relying on a changeup may also result in fewer injuries than relying on a slider or another offspeed option, most likely because it causes less stress on the arm.

Additionally, because pitchers have different arm slots and use different grips to throw the pitch, there are many unique changeup "shapes", or movements, around the league. While it is understood that some changeups are more effective at generating grounders than others, I wanted to identify which specific aspects of the pitches are causing this. To do this, I regressed each pitcher's vertical movement, horizontal movement, velocity and the velocity difference between their changeup and their fastest pitch against their ground ball rate.

My results matched up with Harry Pavlidis' 2013 changeup study—vertical movement (rise) and velocity were negatively correlated with ground ball rates, and velocity difference was positively correlated. Horizontal movement was not deemed significant in the regression.

A negative correlation means that as the independent variable (in this case, the specific pitch component) increases, the dependent variable (groundball rate), decreases. This relationship is demonstrated by the following scatterplots.

Here, we can see that as vertical movement or velocity difference increases, groundball rate decreases. On the other hand, velocity is positively correlated, meaning that as velocity increases, groundball rate increases as well.

In addition to identifying the significant components, the regression also gives a formula which can be used to project groundball rates based on the changeup's shape. This can be useful in determining the legitimacy of a pitcher's performance, or in setting expectations for a young pitcher. If you are interested in seeing the "guts" of the formula, or using the formula to generate expected groundball rates for other pitchers, the spreadsheet can be found in my summary of the study here.

Ideally, we would like to be able to create a model to project other result stats, like whiffs, to get a better understanding of the overall effectiveness of the pitch. But, one of the key components in determining whiff rates on a changeup is how often the pitcher throws the pitch. Since we are trying to project outcomes using a small sample of pitches, looking for whiff rates would mean that we'd only arrive at meaningful projections for a few of the pitchers on the roster.

So even though this yields insight on only a piece of a piece of a pitcher's arsenal, I decided to apply the groundball model to the Rays to try to get a feel for who in the system had the best and worst changeups in terms of generating ground balls. I used pitchers who had thrown at least five changeups that were tracked by PITCHf/x during either this season or last season. It's important to note that these groundball percentages are only for the pitchers' changeups, not their overall groundball rate. Here are the results:

Player	Horizontal Movement	Vertical Movement	Velocity	Velocity Difference	Expected Groundball Rate
Ronald Belisario	5.31	0.96	88.02	7.36	59.82%
Matt Andriese	1.29	0.76	85.00	7.38	58.65%
Colton Reavis	10.02	2.99	86.96	7.63	55.16%
Matt Buschmann	9.64	3.57	84.99	6.36	54.34%
Neil Wagner	4.61	2.58	86.07	10.03	53.08%
Nathan Karns	7.34	5.31	86.84	6.18	52.15%
Jeff Beliveau	11.55	4.28	81.13	6.68	50.72%
Erasmo Ramirez	8.04	2.47	81.30	10.43	50.48%
Brad Boxberger	8.41	1.54	80.65	12.76	49.58%
Kevin Jepsen	8.46	6.67	87.36	7.93	48.07%
Chris Archer	7.92	6.46	87.01	9.26	46.96%
Matt Lollis	9.37	7.26	86.11	7.44	46.81%
Zach Cooper	6.48	6.44	84.39	8.33	46.61%
Ernesto Frieri	6.67	6.95	84.74	7.97	46.18%
Kirby Yates	8.54	5.61	82.50	9.77	45.78%
Alex Colome	4.41	7.11	85.65	9.41	44.89%
Andrew Bellatti	4.68	8.05	85.86	8.17	44.45%
Jamie Schultz	6.91	8.85	86.08	7.99	43.22%
Everett Teaford	8.21	7.27	81.00	9.49	42.16%

A few things immediately stand out. Many of the Rays pitchers feature changeups that generate groundballs close to or above the league average rate (48%). It is unclear if this is an organizational philosophy or just by chance, but it is something worth noting.

With only 0.76 inches of vertical movement, Matt Andriese's changeup has one of the best shapes for groundballs in the organization. FanGraphs' Kiley McDaniel agrees that it is an effective pitch, grading it at a 50, which was tied for highest of all pitches in Andriese's arsenal. But, he has rarely used this pitch, throwing it only 12.57% of the time. I was interested to see if Andriese might be hiding this pitch, with the belief that if he threw it more, he would expect worse results.

To examine this, I looked to see if how often a changeup is thrown is significant in determining groundball rates on that pitch. The regressions concluded that it was significant, but positively correlated, meaning that if Andriese throws this pitch more, we would expect him to get more groundballs. There's still a chance that Andriese doesn't have confidence in this pitch and isn't throwing it often in an effort to protect it. But, based on its shape and McDaniel's grades, it seems like he should try to throw it more.

While Chris Archer's fastball and slider are considered great pitches, he had struggled to find a third offering. He had toyed with a changeup in past seasons, but it hadn't been good enough for him to throw it consistently. But, toward the end of last season, Archer started to throw his changeup more, and it has developed into an effective pitch.  If we look at graphs of its horizontal and vertical movement over the past three seasons, we can clearly see its evolution.

Over time, we see Archer gain more horizontal movement and more vertical "drop," making it a more effective pitch. Archer's current changeup has a shape that suggests it would be around league average in generating ground balls, and helps give Archer a third "band" of velocity to keep hitters off balance.

Overall, this model gives us only a sliver of the components to any pitchers' arsenal. But, it does help us understand why a pitcher gets certain results, and what we can expect from a pitcher going forward. The Rays pitchers generally seem to have a solid bunch of changeups for inducing ground balls, and it is reassuring to see that some of the original "question marks" that were pushed into larger roles, like Andriese and Karns, have solid changeup shapes that can help them build toward success.

Statistics are from BrooksBaseball.net and Baseball Prospectus.