Command and approach:
It’s very easy to go to Trip Somers’ pitch f/x tool to see that in 2010, Phil threw his curve for a strike 54% of the time, and to conclude that because this percentage is lower than the league average of 58% that Phil does not command his curve well. However, this kind of cursory analysis relies on one big assumption: this assumes that Phil wanted to throw his curve in the zone.
Here is where Hughes threw his curve, compared to the locations of a random sample of curves thrown by right handers in 2010:
This graph is from the catcher’s perspective, so the right side is near left handed batters and the left side is near right handed batters. Blue represents locations where Phil threw his curve more than the average right handed curveball, and red represents where Phil threw his curve less than the average right handed curveball. The box is the strikezone. You can click on the graph (and all following graphs) to enlarge.
As you can see, Phil uses a different approach than most right handers. Most of his curves are thrown early in the count to left handed batters — he often tries to “backdoor” the pitch in an attempt to get a called strike. This explains why Phil’s strike percentage is low; he simply doesn’t try to attack the zone with the pitch. He mainly concerns himself with nibbling the outside corner. It appears then that a more advanced method is necessary to try and assess the quality of Phil’s curveball command. One such method is principal component analysis, which has been used to quantify curveball command here. What this tells us is Phil’s ability to throw his curve where he wants to throw his curve, regardless of whether or not this location is the right decision. I was able to duplicate the method used by the author in that article, and I found that Phil’s curve had a lateral command of .60, a value which is nearly equivalent to that of the wicked curveball tossed by Adam Wainright. The pitch also had a upright command of 1.10, a value which would place Phil’s curve between that of Matt Cain and Jon Lester. Pretty good company right there. However, you need to take these values with a grain a salt as this method is not flawless and I am not a professional statistician.
It’s difficult to conclusively state whether or not this tentative approach is optimal. On one hand, Phil is missing an opportunity to get ahead of batters by being so careful, which may cause Phil to fall behind in counts. On the other hand, as shown earlier, when Phil’s curve gets put into play (including homeruns), it goes for a hit nearly 40% of the time. To attempt to quantify the quality of his location and thus his approach, I created a model that predicts the run value of the league average curveball (thrown by a right hander) based on pitch location. The results predicted that Phil’s curve saves 0.24 runs for every 100 thrown (compared to my sample), which is pretty close to average. In other words, if you replaced Phil with the average right hander and he threw curveballs to the exact same locations that Phil threw to in 2010, you would get a neutral outcome.
Phil’s curve has outstanding movement:
pfx_x =horizontal movement compared to a ball thrown without spin (inches)
pfx_z = vertical movement compared to a ball thrown without spin (inches)
vxf = horizontal velocity from the batter’s perspective
vzf= vertical velocity from the batter’s perspective
For every 100 curveballs that exist, Phil’s curve has more vertical movement than 85 and more horizontal movement than 60. It has a tremendous drop and a great sweeping motion. However, this is just movement compared to a ball thrown without spin. It is more useful to know how the pitch moves from the perspective of the batter. To learn about this, please read this piece of research. Using these vxf and vzf values, Phil’s curve does not come out quite as well, though it still seems to be a nasty pitch. While the pitch does have below average velocity, this is not something that is very important to the effectiveness of curveballs.
In fact, one could say that Phil’s curve almost has too much movement…
The above graph shows the pitch flights of Phil’s fastball and curveball from a side view. The horizontal axis indicates depth (distance from home plate) and the vertical axis indicates height of the baseball (above the ground). The red indicates curveball (CU) and turquoise indicates fourseam (FF). The right side of the graph is the release point and the left side is home plate. The curveball stops a little short because I only charted the flights for the times up until when the fourseam reached home plate. Because Phil’s curve is slower than his fastball, it is about 10 feet from home plate at the same time his fastball reaches the plate.
The big “hump” of the curve should stand out. These “humps” are very common, but Phil’s is much more pronounced than most. What’s also important here is the locations of these two pitches where I have placed two filled circles. You should see these two dots above the 45 ft. marker (remember to click on the graph to enlarge). These dots indicate when the time is .075 seconds after the pitcher has released the ball. Robert Adair (author of The Physics of Baseball) found that .075 seconds is when the batter needs to decide whether or not to swing, so this is a crucial moment. If the hump is too large, the batter is going to recognize breaking ball and layoff the pitch. This explains why the swing rate on Phil’s curve is much less than average.
Another way to look at this is a motion plot:
In this motion visualization, the blue circle represents the flight path of the curve and the red circle represents the flight path of the four seam. Size of the circles indicates depth; circles are largest at the release point and get smaller as they approach home plate. The reason the curve never falls below the four seam is because I have stopped the time when the four seam reaches home plate.
As you can see, Phil has to throw his curve slightly upward to compensate for the crazy amount of downward movement the pitch gets. Meanwhile his fastball dashes to home plate on a pretty straight path. This creates a pretty large difference between the flights of the two pitches. However, you may be aware that there are other effective curveballs in the league that are both slow and have a lot of movement. Why are these pitches effective but Phil’s is not? Well, there are few reasons. These pitchers may have more deceptive deliveries, a lower release point for their curves, or they are just better at setting up the pitch. A combination of the three or some other factor is likely.
After breaking down Phil’s curve, we can see that he has a decent approach, good command, and wicked movement when throwing his curve. The only problem seems to be that the curve is not very deceptive. Batters are able to recognize the pitch and either lay off it or hit it hard, making the pitch very ineffective. This also explains why Phil’s curve used to be effective; when he first came up he threw a more traditional type of curveball (not knuckle curve) that had less movement than his curve does now. This made the pitch harder to pick up, which made it effective. It is also possible that hitters in the minor leagues are not nearly as adept at recognizing breaking balls as major leaguers, which would allow Phil to throw curve after curve without worrying about deception. However, there are some caveats. I only looked at 2010 data because he has changed the way he has thrown his curve a few times prior to last year. This has the unfortunate effect of making the sample small. It is also entirely possible that he received a bit of anger from the BABIP gods last year; the curve had a BABIP of .366 last year compared to the league average of .290. Based on everything we know about pitchers, it is likely that Phil’s curve has a lower BABIP this year. It is very possible that Phil learns how to better disguise and set up his curveball in the future, which could lead Phil to become the ace that we all thought he would be [a return of his fastball would be nice too].
*This is slugging percentage on balls in play, or non-k slugging. The league average here (~.500) is higher than regular slugging percentage because strikeouts are not included.
*Information from Fangraphs was used in this post, which can be found here for Hughes and here for Burnett. Information from Trip Somers’ pitch f/x tool was also used.
*Pitch f/x data is from Darrel Zimmerman’s pbp2 database
*Run value mysql code from Ricky Zanker
*The deception part of this post borrows heavily from ideas put forth by Josh Kalk. Earlier parts of this post use the fantastic research of Max Marchi, Mike Fast, and Matt Lentzner.