CRPerry13 stands on the backs of giants and applies their knowledge to investigate why Astros pitcher Philip Humber's success rate declined from 2011 to 2012.
A couple weeks ago, the Astros claimed Right-Handed starting pitcher Philip Humber after he was placed on waivers by the Chicago White Sox. Humber was drafted with the 3rd overall pick of the 2004 amateur (Rule 4) draft by the New York Mets, and has since experienced varying levels of success in the major and minor leagues. The highlight of his young career to date (he is 29 years old, but has pitched only 316 major league innings) is undoubtedly the perfect game he threw against the Seattle Mariners in 2012. However, his statistical peak came during the 2011 season, when he posted an ERA of 3.75 (3.58 FIP) in 26 starts.
Humber's first-round pedigree and demonstrated major league success are likely reasons why the Astros claimed him. Unfortunately, his performance in 2012 was the major reason why he was available on waivers in the first place. In 2012, Humber posted an ERA of 6.44 (5.77 FIP). In fact, by FIP (Fielding Independent Pitching - a measurement of runs allowed, normalized to remove the effects of defense behind the pitcher), Humber was the worst pitcher in 2012 out of the 142 pitchers who threw more than 100 innings.
A week ago, Alex Kienholz over at Beyond the Box Score (an SB Nation site) posted an article exploring whether or not Humber can regain his 2011 form. While I could quibble about a couple irrelevant details, there is too much nitpicking between analysts online. I want to focus on his major point and some of the topics he broached that I think are important, primarily about Humber's control and the affect that has on his results. I started researching my post a few weeks ago, and now Kienholz' work allows me to skip over some my original points and dig deeper; it is worth reading his article.
If you've read many of my posts before, you probably know that I am more interested in what did happen, instead of what should have happened. Many sabermetric concepts du jour are concerned with explaining how pitchers should have performed - in a neutral offensive environment, with defense removed, with statistical outliers removed, etc. Those concepts are sometimes valuable to predict whether or not a poor performance is an abberation or likely to correct itself just from a statistical standpoint (the cliché is "regress to the mean"). But sometimes, numbers can't tell the entire story (though they can help with the details). I prefer to look at what actually happened to try to understand why what should have happened, didn't happen. Stepping away from a paradigm of platonic idealism is how analysts can apply their mathematical models to the real world to fix problems, rather than just describing them.
I like to start by looking at the outcomes generated on a per-pitch, per-swing, and per-contact basis. In a previous post, I defined (but probably did not invent) a concept called "Positive Outcome Per Pitch" (POPP), and have applied it to hitter success at the plate. I later expanded that to "Positive Outcome Per Swing" (POPS) and "Positive Outcome Per Contact" (POPC). Using data from MLB Advanced Media collected from their Pitch Trax application and measured by Sportsvision's Pitch F/X system, I define a positive outcome as any pitch that yields the result of "In Play (No outs)" and "In Play (Runs)". The same concepts can be applied to pitchers - only those positive outcomes defined for hitters become negative outcomes for the pitchers.
If you're a fan of acronyms, I'm now talking about NOPP, NOPS, and NOPC.
Using Humber's Pitch F/X data from 2011 and 2012, I calculated his NOPP, NOPS, and NOPC. In the interest of preserving the scroll wheel on your mouse, I've included this raw data at the end of the post. It is not necessary to make my point here, and it would just get in the way, much like this sentence has.
Predictably, because of his inflated ERA and FIP, Humber's outcomes were noticeably worse in 2012 in almost all cases. As Kienholz hints at, Humber's success rates were particularly worse with his slow pitches, particularly his curveball and changeup. Over one out of every five curveballs or changeups that his opponent made contact with yielded a negative outcome - either a new baserunner without adding an out, or a run scored against Humber.
The chart below shows the difference between 2011 and 2012 in terms of negative outcomes:
The good news here is the comparatively unchanged success rates that Humber posted with his Four-Seam Fastball (FF). Given that most pitchers (and Humber is no exception) throw their four-seamer a majority of the time, this is something he can lean on while he attempts to correct his issues with his other pitches.
The Two-Seam Fastball, possibly what Texas Leaguers and Brooks Baseball are classifying as a "sinker" in 2011 and not classifying at all in 2012, shows a dramatic increase in success in 2012. While I am dubious about the accuracy of this pitch identification, if this data is correct then Humber should focus on throwing more two-seam (or sinker) fastballs in 2013, and use that as one of his primary weapons. Predicably, this pitch is most effective against Right Handed Hitters, as it has the ability to drive in towards the hitter. If thrown on the outside of the zone, the pitch will appear to be a ball outside, but its movement will make it tail into the zone for a strike. Often, this causes weak contact because the hitter is swinging late in an effort to foul the pitch off. If thrown on the inside of the zone, the ball will break inward, causing weak contact against the narrow part of the bat.
The bad news is that Humber's slow pitches yielded terrible results compared to 2011. While his NOPP rate has been consistent between seasons, his NOPS and NOPC have skyrocketed. The selectivity batters are showing is reflected in those numbers, as well as in his increase in walk rate (9.5 BB% in 2012 compared to 6.1% in 2011).
Several things can affect a pitcher's rate of success with his pitches. Some of those are quantifiable and some aren't. The non-measurable factors include nagging injuries, psychological state, tipping pitches, and changes in pitching mechanics. Those factors can result in measurable changes in things like pitch velocity and ball spin, which in turn can affect pitch movement, location, and consistency.
Pitch Release Point / Mechanics
I am no baseball scout. Looking at video can tell me if a pitch is good or bad based only by the results generated and by speeds posted by the television crew. I do not know how to identify mechanical flaws, changes in the pitch delivery over time, the likelihood that a delivery will be more or less likely to cause fatigue-related injury, and I certainly don't have access to player medical records.
But Pitch F/X can show us a few things. Namely, it can show us where the pitcher is releasing the ball as viewed by the catcher. Sadly, it does not yet tell us the distance at which the ball is released. (Though the center of the mound is 60' 6" away from the corner of Home Plate, Pitch F/X currently assumes a pitch release point of 50 feet from the plate, because that's the point where it begins measuring the initial pitch trajectory. Hopefully, they address that soon.)
In Humber's case, Pitch F/X shows that the average X-Z location of the beginning of the pitch trajectory has changed between 2011 and 2012. In the chart below, the data points are the average initial x and z pitch locations of each of his pitch types (FF, FT, CH, CU, and SL). The triangles are the average initial measured pitch location for the seasons for all pitches.
(click to enlarge)
The data shows that the initial measured pitch location has changed by two inches over the past season. Two inches does not seem like much to those of us who are seventy-five inches tall, but that distance is almost the width of a baseball. Factoring in unmeasured variations in the Y-direction (releasing the ball closer or further from the plate), the actual release location may be significantly different.
This suggests that Humber's mechanics may have changed since his successful 2011 campaign. Without looking at video, it is impossible to ascertain what those changes were or if they were the definite cause of his increased percentage of negative outcomes in 2012, but I think it likely.
One of the factors that can be affected by changed mechanics is the velocity/speed at which the pitch is thrown. Pitch F/X measures the speed of the pitch at 50 feet away from the corner of home plate (the same location as the beginning of their measured trajectory). The arm angle at which a pitch is thrown, the initial actual distance from the plate of release, and brute force can all affect the measured initial velocity.
This chart shows the changes in average initial velocity for Humber's pitches between 2011 and 2012:
Most of these changes in velocity can be dismissed as negligible, but lo-and-behold, the speeds of Humber's curveball and changeup have changed the most significantly. This further hints that he is struggling with those two pitches. In 2012, Humber's changeup averaged a higher speed than his slider. A changeup is not supposed to be fast - it is supposed to be slow. An average increase of over 1 mph is significant for this pitch and can have a major effect on its success. To compare, think of how many analysts make a huge deal over a loss in average fastball velocity of one mph over the course of a season, and the effect it has on a pitcher. Even the change in average velocity of the curveball of about .4 mph can have an impact on that pitch's effectiveness.
It's possible that Humber is throwing his change with more oomph, but its more likely that the difference is the mechanical change that I implied above. He could be releasing the change closer to the batter or releasing it with a different spin that affects its trajectory, and therefore the forces acting on the pitch that slow it down.
Pitch Movement and "Control"
That leads me into possibly the most interesting aspect of my recent studies. Yesterday, I read a great article by Alan M. Nathan of the University of Illinois on how pitch movement is affected by the forces of gravity, drag, and the force generated by the spin of the baseball (called magnus force). In it, he shows how removing the effects of gravity and drag allows one to isolate the ball spin that causes "break" (deviation of the pitch trajectory from a straight line). Essentially, this force is what determines what we typically call pitch movement, since gravity and drag are comparatively constant pitch-to-pitch.
This illustration shows how I am defining pitch movement:
(click to enlarge)
One of the things that can affect pitch trajectory is the speed and angle of the ball spin during its flight to Home Plate. Pitch F/X measures these things as well, though I do not know the percent error on these measurements. Humber's spin data shows little variation in his fastball and slider (catching a theme here?) but comparatively large changes in spin for his curveball and changeup.
Changes in ball spin also point to a change in mechanics because if a pitcher's mechanics and release point stayed perfectly consistent, there should be little variation in the spin of the ball as it traverses the infield towards the catcher's mitt. This is further evidence that Humber struggled with those two pitches in 2012, contributing to his lack in success.
Because of the magnus force on the ball generated by the spin, it is possible to see the affect this change has on the break and overall movement of Humber's pitches. Mr. Nathan provides a great worksheet at the end of his article to correct the pitch movement calculations provided by the "break" field of Pitch F/X, which does not account for drag. This allowed me to look at Humber's true pitch movement while only considering the affect of gravity and the magnus force.
What I found was that Humber's fastballs, curveball, and slider showed little variation in the size of their breaks between 2011 and 2012. However, the changeup exhibited a large difference in average movement.
Another interesting piece of information that can be gleaned from this data is how consistent Humber has been with repeating the movement of his pitches. This consistency can be a major part of a pitcher's "control". Control is an interesting concept, and to define it as the ability of a pitcher to throw strikes is a gross oversimplification. Environment and health both play an effect on control, but so does the ability to consistently repeat a throwing motion. A consistent throwing motion can generate a pitch movement (spin+velocity=trajectory) that is so repeatable that the pitcher will always know where the trajectory will cause the ball to cross the plate if he executes perfectly. From the movement analysis that I studied from Mr. Nathan, I used the standard deviation of the X- and Z- components of the change-in-location vectors (dX and dZ) to look at how Humber has been able to repeat his pitch movement.
This data is the most damning of all to his breaking pitches. While I'm no statistician, what I do know is that the higher the standard deviation in the dataset representing the dX and dZ components of his pitch trajectories, the more variation in the pitch trajectory between pitches. In plain English, in 2012 Humber was far less able to repeat the motion that generated a consistent break (and thereby movement, and thereby location that the ball crosses the plate, and thereby "good control") for his breaking pitches than he was in 2011.
The interesting thing is, Humber was able to execute better control of the trajectory of both of his fastballs (whether you want to classify the two-seam as a sinker or not), and this undoubtedly led to his improvement in NOPP, NOPS, and NOPC for those pitches in 2012. This is shown by the negative standard deviation of pitch movement on those pitches.
The data in this post shows that Humber's struggles in 2012 primarily revolved around his inability to consistently throw his breaking pitches. This is most likely the result of a change in pitching mechanics, partially evidenced by a 2 inch change in release location, the initial velocity of those pitches, and ball spin as measured by Pitch F/X.
Humber was arguably more successful with his fastballs in 2012 than he was in 2011. One path towards regaining good results could be to spend practice time correcting the issues with his breaking pitches, while relying on the strength of his fastballs during games. Choosing to throw breaking pitches primarily in pitcher-favorable counts (a piece of data that I declined to look at in this post but is relevant to the conversation) could maximize the effectiveness of those weaker pitches while improving his own confidence and strengthening the muscle memory that will allow the correct mechanics to be repeatable.
There seems to be no answer as to whether or not Humber should perform better in 2013, since popular metrics such as BABIP were within normal ranges last season. But, it is clear that Humber can perform better if he addresses the issues that led to reduced success with his breaking pitches.
* * *
I'd like to extend my admiration to those researchers whose work educated me enough to write this post, namely through published internet articles by Mike Fast, Alan Nathan, Alex Kienholz, Tom Tango, and the masterminds behind statistical evaluation sites like Fangraphs. Without the ability to build on their work, my posts would be relegated to asinine opinions based on gut reaction and mass media noise.
(click for larger image)