A conversation on the mound after a curveball ended up being hit for a home run.
Nuke LaLoosh: [softly, infuriated] I held it like an egg.
Crash Davis: Yeah, and he scrambled the son of a bitch.
–Bull Durham 1988
The curveball has been around almost as long as baseball yet there still is plenty of mystery around the pitch. Curveballs are all about deception. Ideally, a pitcher would like to throw the pitch with the same arm speed at the same release point only to have the bottom drop out at the last instant leaving the batter wondering happened. When the pitch loses its deception it can be hit a long way as Nuke found out. While I doubt many catchers in the majors are letting batters know that a curveball is coming there still are plenty of ways for the batter to sniff out the pitch.
If you have been reading your 2008 Hardball Times Annual (still available I might add) or following John Walsh’s recent articles you already know that John has worked out a very strong metric for measuring the quality of a pitch using the PITCHf/x data which he calls runs100. If you aren’t familiar with runs100 I encourage you to read John’s excellent article. In any case, here is a simple explanation, runs100 is the value, in runs, of a certain pitch type (fastball, change up, curveball, etc…) for a certain pitcher per 100 pitches.
While runs100 does an excellent job of finding which pitchers throw the best curveballs it doesn’t say why these curveballs are so good. As we have already seen, a lot goes in to throwing a good curveball and it would be nice if we could determine what attributes are the most important and what attributes are less important. This is what I would like to look at today with curveballs and with other pitches in my next few articles. So the plan will be to take certain variables gathered by PITCHf/x and then find how well they correlate to curveball success. First though, I need to introduce a few new variables to the mix.
The curveball hump
The curveball hump is a term you might have heard of from reading an article that quoted a scout or maybe your local broadcaster mentioned it or maybe you picked it up from reading hardballtimes articles. Basically, when someone says a pitcher’s curveball has a hump what they mean is the curveball rises up in comparison to a fastball on the way to home plate. Sometimes this is referred to as a pitcher’s curveball and fastball aren’t on the same plane. You can see this visually if you stand to the side close to where the base coaches stand and then track the pitch. Here is an averaged curveball and fastball from Ted Lilly in 2007.
I have chosen to beginning tracking the ball at 55 feet from the front of home plate and that is labeled as 0 in the horizontal direction here. While the mound is at 60’6″ I don’t want to begin tracking until I know for sure the ball has left the pitcher’s hand so I use 55 feet. Lilly throws a 12 to 6 big breaking curveball. Because his curve has so much downward movement compared to a ball thrown without spin it must be thrown with a higher height in the middle or it would just continually bounce before reaching home plate. The line between the averaged fastball and curve represents the distance where the separation, or hump, is the largest. The two tick marks to the left of that are how far the ball has traveled in 0.075 seconds. This is the time Bob Adair lists as the information gathering time for the batter in The Physics of Baseball (p 38-46). Because the curveball hasn’t traveled as far as the fastball it’s tick mark is closer to the mound than the fastball.
If a pitcher wants to try to conceal his curveball and throw it closer to the plane of the fastball he has a few options. He can reduce the vertical movement of the pitch or he can lower his release point or both. Dan Haren is a good example of this.
Haren’s release point for his curveball is nearly half a foot lower than his fastball and his curve is much more of a slurve than Lilly’s curveball. In fact, Haren generates more horizontal movement with his curve than he does vertical movement compared to a ball thrown without spin. The result however is his curve and fastballs are on virtually the same plane during it’s flight. So which of these two approaches is better? Now it is time to run some correlations which will include the maximum height of the hump, the horizontal distance where this maximum occurs, and the height of the hump at 0.075 seconds or Adair’s hump.
In addition to those three variables I will also be testing seven others. Those will be the difference in release point between the fastball and curveball, the speed of the curveball, the speed of the curveball relative to the fastball, the vertical movement, the vertical movement relative to the fastball, the horizontal movement, and the horizontal movement relative to the fastball. Here when I use the term movement you should read that as the movement of the ball compared to a ball thrown without spin. My sample is the 33 pitchers who threw at least 200 curveballs in 2007 according to my pitch classification algorithm. I am using simple Pearson correlations for this study and all variables will be correlated to the runs100 for that pitcher.
None of the variables correlated better than medium correlation (0.3 to 0.5) to curveball success and they were on the low end of that as well. At first I was a little disappointed by that, but then I realized if one of these variables strongly correlated then that would have been discovered already, most likely first by scouts, and then disseminated such that most if not all major leaguers would throw a certain way. The variable that correlated the most was indeed the horizontal distance where the maximum hump occurred followed by the height of the hump and then the speed difference between the fastball and curveball. In fact, all of the variables that used the difference between the fastball and curveball had higher correlations than the variables from the curveballs alone. I guess this is information that Barry Zito has already discovered this year.
So how large of a hump and where the hump occurs (closer to the batter is better) appear to be the most important things when throwing a curve. Release point does show a small correlation so if possible a pitcher would like to release his pitches at the same point but if that is producing a large hump the pitcher is probably better off lowering his release point on his curve if it doesn’t affect his control. Vertical movement was right on the boarder of a small (0.1 to 0.3) and medium correlation but horizontal movement was uncorrelated (0.0 to 0.1). This implies that a 12 to 6 hammer outperforms a three quarters slurve because as your arm angle drops the more horizontal movement gets imparted to the ball. Also, the hump at 0.075 seconds, or Adair’s hump, is uncorrelated to curveball success. Maybe this means even if a batter gets fooled in the information gathering stage he can still adjust and maybe check his swing if the pitch produces a high hump later on.
I strongly believe that PITCHf/x data is the future of player analysis. Not only can we learn what pitchers throw effective pitches but we should also be able to tell why those pitches are effective. Once we know why we can do a better job of predicting future success not only for players in the majors but also players in the minors or even in college. This data can also be used to determine what might be going wrong with a pitcher. For instance, Jason Isringhausen is off to a poor start this year. We can try to shed some light on why by looking at a plot of his pitches from 2007 and this year to see if there is a difference.
In 2007 Isringhausen’s release points were practically on top of each other while still not producing a huge hump and resulting in large vertical movement. What about 2008?
We should be a bit careful looking at this data as it isn’t corrected yet. That could explain why it appears Isringhausen’s overall release point has risen. Even if we do adjust his release point down it is clear Isringhauen’s curve is now being released from a higher point than his fastball. This in turn has caused a larger hump (though slightly later) as well. He now is producing more horizontal movement than in 2007 and the end result is his curve and fastball are ending up very close together when the ball enters the hitting zone. Isringhausen throws his curve about 20% of the time and nearly 50% of the time in 0-2 and 1-2 counts. His strike rate has fallen over one batter per nine innings and it looks like his curve might be to blame for that. What this data can’t yet tell us is if this is just a mechanic issue with Isringhausen or is this just regular aging. Because most of the difference seems to be in the release point which spills over to the added height on his hump my guess is it is something that is fixable and he certainly is in good hands with Dave Duncan as his pitching coach. It will be interesting to track Isringhausen throughout the rest of the year to see if he can regain his 2007 curve or if that pitch has been lost.
References & Resources
I’d like to thank Sal Baxamusa for some key research help and Bob Adair for his wonderful book.