Kyle Hendricks and Location-Based Contact Management

This month last year, Connor Kurcon of Six Man Rotation set out to quantify the location aspect of command (or “LRP”). By establishing an accounting system that credited and debited pitchers for changes in ball-strike counts based on the attack zone of and hitter’s disposition (take? swing? ball in play?) for every pitch, he effectively created an alternative to Pitch Value (PVal) that rewards optimal movement through ball-strike counts but with much more pitcher and hitter context.

His findings are as you’d expect: Jacob deGrom and Justin Verlander lead the pack, with Gerrit Cole, Max Scherzer, and Clayton Kershaw not far behind. Other budding aces like Jack Flaherty and Mike Clevinger pepper the list, and some pleasant surprises (such as Brendan McKay, Caleb Smith, and, for those still thirsting, Jake Odorizzi) are scattered throughout as well. Out of the bullpen, newly anointed relief ace Nick Anderson led the pack followed by the underrated Emilio Pagán, breakout reliever Giovanny Gallegos, and others.

Near the end of his post, Kurcon includes a subhead dedicated to Kyle Hendricks where he highlights how Hendricks, widely respected as a command artist, fares lukewarmly by measure of LRP. He then reminds us “LRP doesn’t paint the full picture of command.” True that.

Fortunately, Kurcon has left the door open for me to tie up loose ends with find Gs I’ve been meaning to write up for a couple of months now. Never fear, Hendricks is the command artist we know and love — it’s just that he relies heavily on incurring contact in optimal pitch locations. It is a needle very few pitchers can thread, but Hendricks does it masterfully.

Methodology (gross math stuff, feel free to skip)

Back in July, I wrote about launch angle and what pitchers can and cannot control. In the post, I used a nonlinear regression to establish that pitch location influences launch angle. It was cool! If not dense and confusing for many. As noted above, feel free to skip past this if you’re just lookin’ for the goods.

I use the same framework here. Launch angle directly affects weighted on-base average on contact (wOBAcon). Launch angle relative to pitch location, therefore, should directly affect wOBAcon relative to pitch location as well (in other words, the results there and here are invariably related), justifying the same fundamental model structure.

To set up everything, I calculated the average weighted on-base average on contact (wOBAcon) for every [pitch_x, pitch_z] set of pitch coordinates (horizontal location and height, respectively). I also flagged fastballs (binary variable: 1 for four-seamer/two-seamer/sinker/cutter, 0 for not), which, in my research on launch angles, were shown to have a more dramatic effect on launch angle at any given pitch location. To make the regression nonlinear, I included squared terms for pitch_x and pitch_z. Launch angle did not behave linearly; as a pitch strays from the center of the plate, launch angle tends to decrease. It stands to reason, then, that contact management bears a nonlinear relationship with pitch location as well. Lastly, I omit certain sets of pitch coordinates to which pitchers rarely throw to reduce noise.

In what I’m calling an alternative framework, instead of flagging all fastballs, I flag all pitches thrown less than 91 mph (1 for velocities up to but not including 91 mph, 0 for all else). This caters specifically to Hendricks but carries implications beyond him (possibly to other low-velocity command artists such as Zack Greinke) because Hendricks’ fastball differs greatly from other fastballs; as such, this simple binary (fastball or not) might misconstrue the (in)effectiveness of Hendricks’ sinker. I want to test this hypothesis, that his lower-than-average velocity actually benefits him because he combines it with pinpoint command.

(To be 100% clear: the 91-mph threshold is completely arbitrary and absolutely open to revision/debate. I just thought to myself, Off the top of my head, what is Hendricks’, like, max velo? Probably 91.)

The two equations are as follows…

Traditional: wOBAcon = β * pitch_x + β * pitch_z + β * pitch_x² + β * pitch_z² + β * fastball + ε

• Adjusted r² = 0.70

Alternative: wOBAcon = β * pitch_x + β * pitch_z + β * pitch_x² + β * pitch_z² + β * less_than_91 + ε

• Adjusted r² = 0.72 (interesting — it’s better!)

Both models are quite strong but fall short of my models for launch angle. That should be reasonably expected. You can influence launch angle (and exit velocity), but it’s out of your hands, so to speak, after that. Moreover, small nuances in a batted ball event could separate a single from a double, a double from a triple, a long fly ball out from a home run. wOBAcon, while an effective catch-all metric to measure production (hitters) or damage allowed (pitchers), fails to capture the probabilistic nuance — aka, random chance — of every batted ball that makes baseball great.

Results

No horsin’ around — here’s how Hendricks ranks in location-based “deserved” wOBAcon annually under the traditional “flag all fastballs” framework:

• 2017: 8th of 127 (min. 300 batted ball events)
• 2018: 5th of 126 (min. 300 BBEs)
• 2019: 24th of 116 (min. 300 BBEs)
• 2020: 15th of 114 (min. 100 BBEs)

And here’s how Hendricks ranks annually under the alternative “flag pitches less than 91 mph” framework:

• 2017: 2nd of 127
• 2018: 3rd of 126
• 2019: 14th of 116
• 2020: 11th of 114

By virtue of where he ranks among his contemporaries, Hendricks’ reduced velocity actually does appear to benefit him! (To be clear, I am reluctant to claim that this, a slower fastball being better at managing contact, is a universal truth. I think velocity is only as effective as location allows it to be. A well-placed 89-mph sinker from Hendricks is gold, but that same pitch grooved down the middle might get torched.)

Here’s a list of pitchers who have cracked the top-25 multiple times since the start of 2017, sorted by number of times (I relaxed the threshold to 250 BBE for this one). These results follow the traditional framework and would inevitably vary under the alternative (and possibly more compelling) framework. However, I had already compiled this list before developing the alternative framework and am too lazy to compile the list again (sorry):

• Four times:
  • Dallas Keuchel (1st in 2017, 13th in 2018, 17th in 2019, 5th in 2020).Probably one of the first pitchers someone thinks of when they hear “contact management.”
• Thrice:
  • Aníbal Sánchez (6th in 2018, 11th in 2019, 20th in 2020). No coincidence he experienced a resurgence in 2018-19, and it’s possible he got unlucky this year.
  • Blake Snell (22nd in 2017, 18th in 2018, 17th in 2020). More evidence that 2019’s 4.29 ERA and .343 batting average on balls in play (BABIP) were outliers.
  • Hyun-Jin Ryu (3rd in 2017, 2nd in 2019, 8th in 2020). He didn’t meet the threshold for 2018, but he would’ve ranked highly if he did (and joined Keuchel in the “four times” club). One of the best in the biz.
  • Jason Vargas (4th in 2017, 4th in 2018, 7th in 2019). Bad pitcher who thrived on elite home run prevention.
  • Kenta Maeda (15th in 2018, 6th in 2019, 2nd in 2020). Finally getting the respect he deserves.
  • Masahiro Tanaka (19th in 2017, 8th in 2018, 16th in 2019). Let’s get this man out of Yankee Stadium and see what he can do!
  • Zack Greinke (5th in 2017, 3rd in 2018, 5th in 2019). Like Hendricks and Keuchel, one of the first names to spring to mind when “contact management” is mentioned.
• Twice:
  • Alex Wood (2nd in 2017, 20th in 2018). I admit, Wood is fantasy kryptonite for me. After this year it’s time to cut bait.
  • Eduardo Rodriguez (23rd in 2018, 4th in 2019). An interesting inclusion here. Here’s to hoping he recovers fully from COVID-19.
  • Gio González (17th in 2017, 14th in 2018). The wheels came off in 2018 but he still held it together despite horrible strikeout and walk rates. His 2017 campaign was legit!
  • Jacob deGrom (15th in 2019, 18th in 2020). I mean.
  • John Means (10th in 2019, 11th in 2020). Super weird season this year, but I think I’m buying his ERA beating his estimators. It’s just a question of if 44 innings this year supersede 155 innings last year. (His best 10-game K-BB% rate never eclipsed 14% last year… it’s 19.9% right now. I’m inclined to buy, and fade the 21.8% HR/FB.)
  • Jon Lester (8th in 2017, 25th in 2019). Fart noise.
  • Marco Gonzales (2nd in 2018, 12th in 2019). He cracks the top-25 under the “alternative” framework. There’s a lot to like here from a deep-league perspective. I’m just concerned his low-3.00s ERA this year inflates his draft price a little too much next year. He’s more likely the 4.00 ERA guy we already know.
  • Tommy Milone (3rd in 2019, 1st in 2020). Uhhhhh…
  • Trevor Richards (21st in 2018, 13th in 2019). Bad pitcher, half-decent home run prevention.
  • Wade LeBlanc (1st in 2018, 1st in 2019). This helps explain the .281 BABIP since 2014. Again, pitch to contact only gets you so far if you can’t strike anyone out.
  • Zach Davies (9th in 2019, 4th in 2020). Maybe his 3.30 ERA the last two years isn’t so lucky after all (although it probably shouldn’t be a full run lower than his 4.35 FIP.) The 39-point difference between his career ERA and FIP checks out on first glance.

It would have behooved me to include a list of bottom-feeders as well. Again, laziness overtakes me, but I will make note of one pitcher in particular: Germán Márquez has consistently underperfomed locationally, except during — you guessed it — his 2018 breakout campaign. If you pitch in Denver you should probably be spotting your pitches much more effectively. It’s wild to me to think Márquez has taken the bad hand Coors Field has dealt him and somehow made it worse. (If I wasn’t clear about this before, know now: these results are agnostic to ballpark and defense.)

Discussion

Like Kurcon’s LRP, location-based analysis is not the be-all and end-all of quantifying pitcher command. But I do think it is a solid foundation for discussing the merits of contact management. We see distinct outliers, both good and bad, that fit our understanding of those pitchers. What’s more, many of those pitchers have shown an ability to repeat their location-based success (or lack thereof) year over year.

And this model isn’t overly rigorous to begin with — it cultivates results using location alone without considering countless other ways to describe a pitch, such as its physical characteristics (movement, release point, etc.). Something more statistically complex undoubtedly would bear even more fruit.

Paul Mammino of RotoFanatic introduced an ERA estimator called Stuff-ERA earlier this year. It also relies on location as one of its foundations and uses it, among other things, to estimate wOBAcon. The models are (as just noted) complex but opaque, making them difficult to judge, but many of the “surprise” names (Vargas, Davies, Milone) are repeated here, as are some of the headliners (Maeda, Snell, Greinke).

ERA estimators like fielding independent pitching (FIP), expected FIP (xFIP), and skill-interactive ERA (SIERA) make fairly broad, often incorrect assumptions about pitcher (in)ability to manage contact. FanGraphs uses FIP is calculate wins above replacement (WAR). If we can quantify contact management more reliably, we can confidently explain why, for example, Hendricks’ career 3.12 ERA is four-tenths of a run lower than his FIP (3.53) — and why Hendricks deserves a few more WAR than he has been credited.

As for this analysis’ incompatibility with LRP: LRP does many things, but it does not specifically measure contact quality allowed. This does! That’s the simple explanation.

As these differences pertain to Hendricks, LRP relies on changing game states to attribute run values to a pitcher. LRP likely admires Hendricks’ pristine walk rates, but fundamentally he is not a strikeout pitcher. He relies prominently on pitching to contact, which is inherently less valuable than racking up strikeouts, no matter how good one is at the former.

Out of necessity, LRP oversimplifies some of its assumptions related to pitch location and the value of balls in play based on generalized attack zones and ball-strike counts. These simplifications are beneficial broadly but likely do Hendricks a disservice. He is not inducing whiffs in the chase or waste areas, nor is he overpowering hitters in the heart of the plate. He lives, literally and idiomatically, on the edge (or, using Statcast’s nomenclature, in the shadows).

Lastly, this exercise heavily condenses the range of estimated values. Since the start of 2017, single-season wOBAcon values (among pitchers who incurred at least 300 BBE) range from .285 to .483, with a mean of .381. Location-based “deserved” wOBAcon finds a similar mean (.384) but estimates a much narrower range (.356 to .405), with a variance roughly one-seventh as large. Such is the nature of these regressed values and of trying to wrangle the discrete values wOBA assigns to batter ball outcomes, which are widely varied. Maybe a probabilistic model (ordered probit?) might better capture the effect I’m looking for (i.e., increased/decreased probability of incurring damage) but for now that’s neither here nor there.

It is better, then, to proceed with the understanding that the tables below are more important in terms of rank and magnitude rather than actual value. (Because, again, location is but a small part of a much larger puzzle.) As I just alluded to, think of it as: a lower location-based “deserved” wOBAcon implies a lower probability of incurring damage, and vice versa. My recommended use: Verify uncommonly low or high BABIPs and HR/FBs, as opposed to taking the estimates at face value. This is helpful especially for players with short track records and also veterans with long track records who see an abrupt shift in their results.

Tables

Because the juiciness of the ball has fluctuated so dramatically year to year, it is difficult to compare results across seasons within a single table. (Actual and “deserved” wOBAcon values are universally higher in 2019 than, say, 2015, which may unfairly favor Player X or discredit Player Y by comparing apples to oranges.) Accordingly, I have provided separate tables for 2019 (full season) and 2020 (through Thursday, September 24).

2019:

2020: