I call it the Mariano Rivera rule. When explaining DIPS – the theory that all pitchers have very similar skill on balls in play – it helps to state the concept in terms of real pitchers. For most Joe Shmoes, we can expect a BABIP close to league average. Perhaps his BABIP won’t be average in a month, or a year, or maybe multiple years, but eventually we can expect it to approach around .290.
But we know there are exceptions. If a pitcher allows very weak contact, he can allow fewer hits than expected on balls in play. And what pitcher allows weaker contact than Mariano Rivera? In over 1200 career innings, he has a BABIP of .262, which is well below the league average. If we assume that this number is an accurate representation of his true BABIP talent level, which it should be at this sample size, then we can reasonably say that any pitcher with a BABIP below or approaching .262 is at least partially lucky. Of course this is not meant to say that Rivera’s BABIP has some sort of asymptotic purpose in nature, just to suggest an easy rule with a high level of accuracy.
Owner of a career .265 BABIP in over 1300 career innings, Matt Cain breaks this rule. While he pitches against easier competition in a more pitcher friendly ballpark, he is also a starter, forced to weave his way through multiple lineup turns with declining stamina. While his DIPS defying ways have been well documented, they remain fascinating.
Consider a visual inspection of his BABIP performance:
This visual depicts Matt Cain’s BABIP performance based on vertical location. The y axis represents the vertical height of the pitch, where the top of the graph is the top of the strikezone and the bottom of the graph is the bottom of the strikezone. The x axis indicates BABIP. The graph is split up into two visuals, one for performance against right handed batters (denoted by the header “R”) and the other for his performance against left handed batters (denoted by the header “L”). Cain is represented by the red line, and a random sample of 25,000 pitches thrown by right handed pitchers in 2011 is represented by the blue line. The gray bands indicate confidence.
The data used here includes all of Matt Cain’s pitches from 2008-2011, which gives us 2536 pitches that were put into play. This time range is arbitrary, but it provides the best balance of PITCHf/x data quality (2007 data is suspect) and sample size. There is a slight penalty that I am giving to Cain here simply for having pitched earlier than 2011 because the average BABIP was a little higher prior to this year, but this will not substantially affect this analysis.
For right handed batters, Matt Cain and the league perform very similarly. Indeed, for 2008-2011 I calculate his BABIP against righties to be .279, which is close to the sample’s BABIP against right handed batters. Of interest here is his performance against left-handed batters, where he significantly outperforms the league, posting a .259 BABIP compared to .295. This means that in the past four years he has had a reverse platoon BABIP split, despite the fact that he gives up more flyballs to right handed batters. Also of note is that his BABIP advantage to left-handed batters is not concentrated in one area, but rather somewhat evenly distributed across the entire vertical spectrum of the strikezone.
Matt Cain and pitch selection
For the 14,000 pitches thrown by Matt Cain during this period, I have Matt Cain throwing the following breakdown of pitches:
CH CU FF FT SL 0.14 0.12 0.56 0.05 0.13
(CH = changeup, CU = curveball, FF = four-seam, FT = two-seam, SL = slider)
These classifications are a combination of the Gameday algorithm output, manual reclassification, and some automated classification using clustering analysis. I should also remark that while his overall pitch selection has been pretty consistent across all four years, at the beginning of the 2010 he started throwing a few less fastballs which he has replaced with changeups – now about 17 percent.
But these overall numbers are misleading; after all, Matt Cain only pitches to right handed batters or left handed batters at one time.
Against righties he throws:
CH CU FF FT SL 0.06 0.12 0.56 0.05 0.21
And against lefties:
CH CU FF FT SL 0.22 0.13 0.56 0.05 0.05
As you can see, Matt Cain makes heavy use of either the slider or the changeup – two pitches known for their relevance to platoon splits – depending on the handedness of the opposition. Throwing more changeups instead of sliders against lefties may seem obvious, but not all pitchers capable of making this adjustment do so, like Daniel Bard who throws too many sliders to lefties despite having a nasty changeup in his arsenal.
Since it is his performance against left-handed batters that is special, here is how his pitches against lefties perform in terms of BABIP, compared to the sample’s performance against lefties:
This graph indicates the difference between Matt Cain’s BABIP and the league’s BABIP to left-handed batters, split up by pitch type. If the difference is “-0.05″, then Cain’s BABIP for that pitches is .05 less than the league average right handed pitcher. For example, Matt Cain’s BABIP on changeups against left handed batters is .236, which is .046 less than the league rate of .282.
Most important here is that he significantly outperforms the league average on both changeups and four-seam fastballs, which compromise 78 percent of his pitches to left-handed batters. Before running into this analysis I also assumed that he pitches backwards much more than average, but that does not appear to be the case. He does throw less fastballs than average in hitters’ counts, but he also throws less fastballs overall than average.
It’s still not quite clear how exactly he is suppressing hits against left-handed batters, but it seems possible that is it related to the frequent usage of his excellent fastball – change combo which may help keep batters – and sabermetricians – guessing.
References and Resources
* PITCHf/x data from MLBAM via Darrel Zimmerman’s pbp2 database and scripts by Joseph Adler/Mike Fast/Darrel Zimmerman