Quantifying the Benefit of Spray Angle to xwOBA

Expected weighted on-base average (xwOBA) is one of Statcast’s most important additions to the Sabermetric sphere. It’s a simple premise — estimate a hitter’s deserved production based, simply, on his combinations of exit velocity (EV) and launch angle (LA) — with robust implications and applications. It’s remarkable how powerful the metric is with just two inputs.

However, the metric is not without its faults (or complaints from those who use it). Its simplicity is beautiful but inherently and knowingly lacking, accounting minimally or not at all for:

1. spray (lateral) angle (touched upon here),
2. a player’s foot speed (discussed more thoroughly here),
3. park factors, and
4. opposing defense.

None of this necessarily serves as an indictment of xwOBA. The number of inputs you include affects the purpose you want it to serve. That is, do you want it to be descriptive or predictive? How about both? Maybe defense shouldn’t be included, then, if we can’t reasonably expect a hitter to face the same caliber of defense each year, something that is out of his control.

The same could possibly be said of spray angle, although I don’t believe its exclusion from xwOBA to be theoretically sound. Its exclusion implies it’s not a skill the way launch angle is a skill, and I’m not sure that’s true. That said, the inclusion of spray angle into xwOBA would make every EV-LA combination that underpins xwOBA quite a bit noisier — something that might actually reduce the effectiveness of xwOBA. But we can see, pretty clearly from this post, that spray angle (and park factors) matter.

While updating my Pitch Leaderboard ahead of the 2020 season, I have been cleaning up Statcast data, including making minor adjustments to xwOBA to account for some of the factors mentioned in the bulleted list above. Here, I want to demonstrate what effect some general binning of spray angle can have on xwOBA on contact (xwOBAcon), the variable portion of xwOBA. (The three elements of xwOBA: strikeouts, walks, and xwOBA on contact. The former two are fixed, while the latter two depend on EV and LA, as previously noted.)

First, I calculated spray angle using colleague Jeff Zimmerman’s equation cited here by Bill Petti. In theory, we have 90 individual degrees, when rounded to the whole number: -45° (left field) to 0° (dead center) to 45° (right field). I binned all launch angles into nine 10-degree ranges, such that all balls in play between -45° and -35° were grouped together, and so on. This method of binning implies some semblance of repeatability (i.e., so-and-so hits it to left-center field 15 percent of the time) as opposed to assuming he will repeat his exact distribution of launch angles (i.e., he hits it at -30° exactly 6% of the time).

An underrated, although fairly intuitive, element of spray angle: its importance depends on the type and quality of batted ball. I have tried to account for that here as well, using “Quality of Contact” (‘estimated_woba_using_speedangle,’ per CSV documentation) and Batted Ball Type (‘bb_type’) to create additional groups. I have consolidated all weak batted balls (“Poorly/Under,” “Poorly/Topped,” and “Poorly/Weak”) into a single “weak” category alongside barrels, solid contact, and flare/burners.

Lastly, I account for hitter handedness.

All told, this establishes a ton of individual bins — 288, to be exact (two types of handedness times nine spray angle bins times four types of batted balls times four batted ball quality distinctions). Fortunately, we have more than 120,000 batted balls in 2019, leaving an average sample size of roughly 400 batted balls per bin. It is not distributed that neatly, but I think it all comes out in the wash. (A bin with a sample size of one — this actually exists, a lone “barreled” ground ball — will have an extremely small effect on the population.)

I’ve gotten this far but, naturally, I’ve buried the lede: how will I quantify the bias caused by xwOBAcon’s lack of spray angle? Simple: calculate the average difference between wOBAcon and xwOBAcon (wOBAcon–xwOBAcon) for each bin and apply it to every batted ball event (BBE) in 2019. xwOBAcon does all the heavy lifting for us — we know what it “thinks” the value of a particular BBE should be, and we can compare how far off its estimates are from what actually occurred over fairly large samples.

As noted, there are nearly 300 bins, so it’s not worth showing all of their individual wOBAcon–xwOBAcon “biases” here. Instead, I will simply list the ones that occurred most frequently in 2019. Note that “spray_direction” below does not coincidence with location; a ground ball hit toward right-center field does not mean it ended up there (in fact, it was probably fielded by the second baseman).

Per the table above, the average xwOBAcon for a ground ball toward right-center field exceeds the average actual wOBAcon by about 70 points — that is, the xwOBAcon value for this particular BBE bin is potentially 70 points too high. Let’s adjust for that!

This is probably why you’re here, beyond the drudgery of my methodology: here is a list of all hitters with 150+ BBE in 2019, with their xwOBAcon pre- and post-spray angle adjustments, sorted descending by biggest changes in xwOBAcon (‘delta’). Basically, a large positive delta suggests that xwOBAcon had been undervaluing so-and-so’s BBEs (by virtue of contact quality), and vice versa.

Above all, remember that the bins used here are somewhat generalized. It’s the nature of the bast. But I imagine something more specific than this would still yield similar results. However, incorporating park factors might change these estimates slightly, too, especially to dead left, center, and right fields (I doubt the gaps matter as much). Consider these estimates below effectively park-neutral.

It’s an extremely good Friday for Jose Ramirez, Willie Calhoun, Matt Olson, and Jeff McNeil fans (me!!!!).

Note: my wOBAcon and xwOBAcon values do not perfectly align with Statcast’s values. This is because I have made small adjustments to the coefficients/assumptions for actual wOBAcon, forced errors to zero, adjusted for nulls, and so on. These small tweaks will ultimately lead to small (generally not meaningful) discrepancies. Just a heads up!