2017-19 Hitter Launch Angles, Revised

Recently I outlined how the installation of Hawk-Eye as Major League Baseball’s tracking and data collection system has shed light on the issue of untracked batted ball events (BBE) in prior years. The issue was first broached by Connor Kurcon, who uses launch angle for his various research and analytical endeavors, including classified run average (CRA), dynamic hard hit rate (DHH%), and TrueHit percentage.

If you’re too lazy to click through, I’ll recap: Because Hawk-Eye tracks more than 99% of BBE, we can use the distribution of launch angles in 2020 to identify the possible launch angles of untracked BBE in previous years. Most likely, untracked BBE converge on the most extreme angles — think -90° and 90°, but with a margin for error such that some BBE as shallow as -40° (for ground balls) or 50° (for pop-ups) might have still gone untracked.

Absent the information available to us now, Tom Tango and the Statcast team devised a method that would impute exit velocity (EV) and launch angle (LA) values that most closely mimic the untracked BBE’s observed outcome by measure of weighted on-base average on contact (wOBAcon). From my observation, Statcast applied roughly half a dozen different launch angle estimates for this purpose, with two in particular used disproportionately: -21° or -20.7° (for ground balls) and 69° (for pop-ups).

Again, absent the data we now have, this was as good an approach as one could reasonably expect. But now we know untracked BBE cluster around the extremes. An imputation of 69° for pop-ups is reasonable, but -21° for grounders might not be extreme enough.

To correct for this issue in the seasons preceding 2020, I adopted an approach I recommended in my original post.

I flagged all untracked BBE, which took a bit of effort because it’s not immediately clear which BBE weren’t tracked! One can’t simply say, “all BBE hit at -21° were not tracked” — that’s absurd. The key is identifying the EV counterpart that appears disproportionately in the data. For -21°, it’s 83 mph (or 82.9 mph before rounding); for 69°, it’s 80 mph; and so on and so forth for other less-common combinations.

Separately, I calculated the average launch angle of all tracked ground balls with launch angles shallower than -40° and pop-ups with launch angles steeper than 50°, by year. Those averages range from roughly -50° to -52° for grounders and hover around 63° for pop-ups.

Then, for all untracked BBE with negative launch angles (-13° or lower), I applied the year-specific average launch angle between -50° and -52°; for all untracked BBE with positive launch angles (39° or higher), I applied the average launch angle of roughly 63°.

[Quick point of clarification: I used launch angle instead of batted ball type (‘bb_type’ in Statcast) for quality control. In this case, all negative launch angles were associated with ground balls (‘ground_ball’), while all positive launch angles were associated with pop-ups (‘popup’), fly balls (‘fly_ball’), and one lonely line drive (‘line_drive’).]

And that’s it! It mostly fixes the problem associated with launch angle tightness captured in this Tweet:

The introduction of HawkEye and increased tracking capabilities might make the analysis of YoY launch angle standard deviation pretty difficult. I had this same issue with DHH%. LA_stddev by year:

2017 – 25.3
2018 – 25.1
2019 – 25.0
2020 – 28.5 pic.twitter.com/ieiiMxBeJM

— Connor Kurcon (@ckurcon) November 12, 2020

Here’s the after:

2017: 30.4°
2018: 29.5°
2019: 29.4°
2020: 28.9°

So, while league-average launch angle consistency is not as consistent (ha) as it once was, it is also much closer to the 2020 league average than it once was, too.

(To resolve this slight discrepancy, one might consider applying a representative distribution of launch angles below -50° and above 50° rather than a single value. But it seems to me a difficult and problematic undertaking: those values would need to be randomly distributed and assigned, possibly according to the hitter’s observed launch angle distribution. Because it’s randomized, it could arguably be re-done any number of times, thereby constantly changing hitters’ average launch angles. For posterity and year-to-year analysis, it’s easier to work with singular values, however oversimplified they may be. Hey, Statcast did it!)

Finally: here’s the goods! The only part you care about. Using this methodology, I re-calculated every hitter’s average launch angle from 2017 through 2019 (min. 150 BBE since the start of 2017)…

…and every hitter’s launch angle tightness.

That’s it! Make of these results what you will.

It’s worth nothing that launch angles decreased pretty much universally. That’s because all of those untracked grounders saw their launch angles decrease by, like, 30° each. That’s no small difference! Whereas untracked pop-ups are roughly the same (although also lower than previously, which contributes to this effect, too). I don’t consider it cause for alarm, but it is interesting to see how league-wide launch angle changes under these assumptions:

Before / After
2017: 11.1° / 9.2°
2018: 11.7° / 10.3°
2019: 12.2° / 10.8°
2020: 12.4° / 12.4°

Some folks might look at those differences and think I’ve made a mistake. Why would the values be so different? It’s true that it’s possible I goofed on my math and coding. More plausibly, maybe the launch angle revolution was (is?) real, and our previous launch angle-related evidence actually undersold the extent/magnitude of it.

Most critically, this correction doesn’t affect the BBE we typically care about most: barrels and other similarly-hard contact. These changes really only affect the contact quality at the opposite end of the spectrum — something, from my observations, that fantasy analysts rarely use in their review of hitters, at least of good hitters. If anything, this may open a bigger can of worms for pitchers, especially when discussing the merits of contact management skills. But that remains to be seen.