I’m pausing my SMOKE research, as the data are telling me I need a more robust “Stuff” metric than I’ve been using to date. That means I need to buckle down and build the pitch quality estimator I’ve been planning to do for a long time, and then circle back to the transparent modelling of SMOKE. I’m dubbing this project FIRE for now, which will stand for Feature Inferred Run Expectancy. Time will tell if either project will be successful.
As I begin building things out, I debated if spray angle, aka which direction the batter hits the ball, is something that the pitcher impacts. Pulled balls have lower launch angles, but higher exit velocities, and depending on the spray angle, the exit velocity / launch angle pairing might produce a different set of probabilities. If pitchers impact spray angle, it would make sense to include it as part of the model. You can read my deep dive on that here:
So I wanted to know if I could find a relationship between a feature of a pitcher’s arsenal, and the spray angle of batted balls. To do this, I hypothesized that pitchers with faster fastballs will have batters slightly behind and pitchers with slower fastballs will have batter slightly ahead, which would mean that slower fastballs would be pulled more.
To test this, I aggregated all pitcher and matchup seasons in the HawkEye era with at least 200 pitches thrown. It turns out, being a left-handed pitcher might be a key factor:
The relationship is matchup dependent (all 4 permutations matter), and it appears that being a left-handed pitcher makes this effect stronger.
Here’s the the same picture, but isolated for the L-R matchup:
The R² for each matchup are as follows:
L-R: 0.09
L-L: 0.095
R-R: 0.02
R-L: 0.025
Keep in mind that this does not include any other information about the pitcher, such as the rest of their arsenal. So let’s see if pitcher spray angle is “sticky” year to year.
Slider Stickyness = R² 0.20 (Min 30 BIP)
Changeup Stickyness = R² 0.15 (Min 30 BIP)
Fastball Stickyness = R² 0.04 (Min 30 BIP)
Sinker Stickyness = R² 0.38 (Min 30 BIP)
Curveball Stickyness = R² 0.07 (Min 30 BIP)
There are 3 things that jump out. First, Fastballs don’t really manage spray angles, as they are not (typically) contact management pitches, and were generally centered on a mean of 0. Second, non-fastballs are generally pulled, which makes sense, as batters are on time for the fastball, but early on the non-fastball. Theoretically, this could measure deception. Third, spray angle on sinkers was the strongest relationship. That’s interesting.
Concluding Thoughts
It is abundantly clear to me that spray angle carries a lot of information about the pitcher, and should be included in FIRE. It may also assist the model in quantifying deception via pitch shape and velocity synergies. This was but a quick, rudimentary analysis, but I think there’s a lot more to dig into here, especially when it comes to quantifying who has the most deceptive secondaries, as in theory, those should have the biggest pull angles.
Its merely an observation but I would say that as a guy who still plays baseball, the RANGE of the spray angle is partially controlled by BOTH the pitcher and batter. The pitcher and catcher work in unison to gang up on the hitter and decide the type of pitch and location and IF the pitcher can execute and hit that spot, then it will control or limit the possibilities of the what the batter can do with the ball even IF he makes solid contact, thus controlling the RANGE of the spray angle. On the other hand, if the pitcher misses his spot, or the type of pitch he chooses does not come out of the hand correctly (looking at you spinning flat cement mixer slider), IF the batter can square up a round ball with a round bat, and has certain gifts in his talent bag (such as being so big and strong, with long levers, he can virtually hook, and thus pull any pitch), why then, the batter can somewhat control the RANGE of his spray angle to a certain extent. --signed just the thoughts of a true lefty batter and pitcher :-)
Lefties: always fascinating!