Contact Management Is and Is Not a Myth

If there were ever a baseball question that keeps me up it night, it’s this: how do the physical properties of pitches affect batted ball outcomes? Many researchers have tackled the subject with varying degrees of success and elucidation. My attempts have focused primarily on a pitch’s ability to generate swinging strikes and ground balls, the first of which used pitcher-level PITCHf/x data while the more recent of which used individual pitch-level Statcast data.

While modeling whiffs and grounders is interesting (and important, too), something strikes me as much more compelling and confounding: the relationship, if any, between a pitch’s physical properties and its batted ball outcomes, whether described as exit velocity, launch angle, or total base-run value allowed, as measured by weighted on-base average (wOBA) or even expected wOBA (xwOBA).

The ability to prove “contact management” as a legitimate and shared pitcher skill has long eluded the Sabermetric community. Assumptions of a league-average batting average on balls in play (BABIP) and, for xFIP, home runs per fly ball (HR/FB) pervade the common ERA estimators (FIP, xFIP, SIERA) we use to gauge talent and assign value. Those assumptions regarding BABIP and HR/FB imply a pitcher’s inability to control them — and there isn’t much evidence to suggest otherwise.

The advent of Statcast, however, has provided baseball researchers an opportunity to explore this concern in much finer detail than ever before — alas, why I’m here. I want to answer the question that keeps me up at night, because I’m not satisfied with the simplicity of xwOBA being predictive of wOBA. What predicts xwOBA, then? What predicts exit velocities and launch angles?(???)

So, here’s my train of thought:

1. Batted ball outcomes are by dictated contact quality.
2. Contact quality is dictated in part by the physical characteristics of the batted ball, such as exit velocity (EV) and launch angle (LA).
3. The physical characteristics of the batted ball are dictated in part by the physical characteristics of the pitch, such as velocity, movement, and release point.
4. Therefore, a batted ball outcome can, in part, be traced back to the physical characteristics of the pitch.

Like my previous attempt to model outcomes (the second link in the first paragraph), I specified multiple regression models using velocity, horizontal movement, vertical movement, horizontal release point, and vertical release point (as well as squared terms and interactions among all of them, in order to capture any nonlinear relationships) as the explanatory variables. Unlike my previous attempt, I excluded spin rate, for reasons related quite simply to a “gut feeling.” My omission of spin rate here could be erroneous, but at least its omission leaves me something to revisit in the future.

My initial attempt to model contact quality relied on pitch-level data and was utterly futile. I had hoped to see robust results given such a massive, granular data set. Instead, I saw the opposite: absolutely no relationship between xwOBA and pitch composition at the pitch-specific level.

Undeterred, I chose instead to roll up the data into “average” pitch shapes by year, pitcher, and pitch type (“2019, Clayton Kershaw, Four-Seam Fastball”). Because, for one reason or another, certain pitches are thrown infrequently, I limited my sample to pitches that allowed at least 100 batted ball events in a season. The resulting sample size was 1,576 year-pitcher-pitch type combinations from 2015 through 2019 and their average velocity, movement, and release point. This produced much better results.

The following table summarizes the adjusted r² values — a measure of the strength of a relationship, from 0 (no correlation) to 1 (perfect correlation) — between pitch characteristics and the dependent variables in the left-most column. I ran two regressions each: one with year fixed effects (“FE”) and one without. Without going too deep into the weeds, year fixed effects help control for idiosyncrasies like the on-again off-again juiced ball.

The bottom row surprised me: a pitch’s properties heavily influence a batted ball’s launch angle. I mean, inherently I know that to be true — why sinkers are known to be ground ball pitches, etc — but I didn’t expect that strong of a relationship. What’s most fascinating to me about this is, while each observation in the data set was grouped by pitch type (“four-seamer,” “slider,” etc.), pitch type itself — the formal label assigned by Statcast — bore no influence on the model results, nor did pitch location. You can explain 68% of the variance in a pitch’s average launch angle allowed without telling the model what kind of pitch it is or where it is thrown. That’s cool!

The correlation between a pitch and its average exit velocity allowed is nothing to sneeze at, either. All things considered, the relationship is, at the very least, what one might call “weakly moderate” (it’s not weak, but it’s not moderate, ya know!).

Venturing beyond exit velocity and launch angle becomes a dubious affair. xwOBA, a composite of exit velocity and launch angle, bears a weaker relationship than either metric individually — a testament to the variance (sometimes called “luck”) associated with compiling optimal combinations of launch angle and exit velocity. And yet optimal EV-LA combinations — the ones that underpin solid contact, barrels, what-have-you — still do not guarantee offensive production, as evidenced by an incredibly weak relationship between pitch properties and wOBA.

For all intents and purposes, the results parallel my train of thought above. A pitch — and, therefore, a pitcher — can dictate to some extent what kind of batted ball might be induced. Beyond that, though, there’s a ton of decay from what should happen to what does happen. It’s that decay that endows “contact management” with its mythos.

Still, there are interesting takeaways here. A pitcher can exert some control specifically over his launch angle given he understands and embraces the complexions of his offerings. That’s not particularly surprising, however, given distinct average launch angles by pitch type (e.g., 5 degrees for knuckle curves, 15 degrees for sliders, 24 degrees for four-seamers).

The elusiveness of pinning down exit velocity suggests to me it’s better for a pitcher to err on the side of shallower launch angles (i.e., ground balls). This aligns with basically everything I’ve ever learned when it comes to rules of thumb regarding pitcher evaluation. Ground balls more frequently turn into hits (higher BABIP) but are significantly less damaging on average (lower ISO); it’s a lower-variance type of batted ball. Again, unsurprising given existing rules of thumb about pitcher evaluation.

So, I guess that’s it. I learned something new, but I didn’t. Technically, contact management (absent any discussion of quality) in and of itself is not a myth — in fact, contact can be legitimately managed to the extent that a pitcher wields his arsenal effectively.

Contact quality management is another issue entirely, however — one that pitchers can manage but only to the extent that Lady Luck will let them. And Lady Luck, well, she’s reluctant to relinquish her influence.