A Needed Update on Launch Angle Tightness by Alex Chamberlain November 23, 2020 As the de facto purveyor of launch angle tightness (or launch angle consistency, both terms that I use interchangeably), it is important I relay to you significant developments related to launch angles in general in 2020. Let the record show I am merely the messenger and Connor Kurcon, whose name graces these pages (or at least my pages) quite often these days, is forever my muse. In 2020, Major League Baseball instituted its new pitch-tracking (and also ball- and player-tracking) system, Hawk-Eye. You can read about its merits here, among them being its alleged ability to “more comprehensively [track] the full flight of the ball”: Furthermore, if the ball leaves the field of view of all 12 cameras (as can happen on high pop-ups and fly balls), the system can then reacquire the ball later in its trajectory as gravity pulls it back into the view of one or more cameras. Hawk-Eye was expected to track more than 99% of all BBE, a significant upgrade from the previous system. Many approached the claim with skepticism. Turns out, the claim may be legit. Kurcon noticed Hawk-Eye all but ruined the year-to-year consistency of launch angle tightness. Consistency is now inconsistent! Specifically, launch angle consistency values (calculated as the standard deviation of launch angle) have nearly universally grown larger in 2020. For the purposes of launch angle consistency, higher is worse, so it gives the appearance (if you’re looking at players individually and not at the larger picture) that a lot of players cratered a bit during the spring season. And it’s not only because of the shortened season, although the season’s length does contribute partly to the discrepancy: 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 – 25.32018 – 25.12019 – 25.02020 – 28.5 pic.twitter.com/ieiiMxBeJM — Connor Kurcon (@ckurcon) November 12, 2020 Let’s rely primarily on the second image. You’ll notice two distinct spikes that disrupt a normal distribution of launch angles. These spikes are associated with balls not tracked by the previous system. The Statcast team needed a process for handling these untracked batted ball events (BBEs). What Tom Tango and his team did, the way I understand it, is use the resulting weighted on-base average on contact (wOBAcon) and batted ball type (ground ball, fly ball, line drive, pop-up) for each untracked ball and assign a launch angle and exit velocity that, on average, would approximate that wOBAcon. This is a perfectly satisfactory approach, especially when there is no data to indicate otherwise and production is the primary consideration. Now, revisiting that second image, see how the red line (2020 Hawk-Eye data) not only skips over the distinct spikes, indicating it has hardly any untracked balls, but also shows a greater density of BBE shallower than -40 degrees (°) and steeper than 50°. In layman’s terms, there are more of these BBE in 2020, and not because of luck — we can intuit Hawk-Eye has caught what the previous system missed. And we can leverage this intuition to revise a now-outdated approach to reconciling the launch angles of untracked BBE. Untracked pop-ups appear to have launch angles north of 50°. Tango et al. assigned a launch angle in the high-60s; meanwhile, the average launch angle on BBE above 50° is roughly 63°. That’s pretty close! Tango couldn’t have known he’d be that close. And while he’s slightly off, one could argue the original assumption need not be changed. On the other hand, untracked ground balls appear to have launch angles south of -40°. Tango et al. assigned a launch angle of roughly -20°; meanwhile, the average launch angle on BBE below 40° is much shallower — roughly -51°. Thirty degrees is a huge difference, especially for an extreme ground ball hitter who might incur a great many of these untracked grounders. At this point, we can assume the more untracked ground balls a hitter has, the better his launch angle tightness would appear to be. Whereas the assigned launch angles of untracked pop-ups might differ from their actual launch angles by maybe 5° on average, it may differ by as much as 30° on average for untracked grounders. That’s… a lot! As an aside: when I think critically about all of this, it makes sense to me that untracked balls actually have the most-extreme launch angles — that is, balls hit, more or less, straight down into the ground or straight up into the sky. If these untracked balls had incurred more traditional launch angles, they probably wouldn’t have gone untracked! Right? Right. So… what does this mean moving forward? My understanding is Tango et al. will revisit their methodology for assigning launch angles (and exit velocities?) now that, as The Dude prophetically said, new sh*t has come to light. Previously, to account for untracked balls, Andrew Perpetua basically proposed using a batted ball’s type (GB/FB/LD/PU) and its tracked coordinates on the field to reverse-engineer its trajectory and specifications. Unfortunately, now we know this approach manifests the same bias that I outline above because it can only approximate launch angle and exit velocity using known values — everything that’s tracked. Until there’s a formal solution from Statcast, I think a reasonable approach would be to apply the average launch angle for all BBE south of -40° (roughly -51°) to untracked ground balls and for all BBE north of 50° (roughly 63°) to untracked pop-ups. While also an imperfect solution, this may mimic the actual distribution of launch angles we observe a little more closely. Or, if you want to circumvent having to fix the data all together: simply index launch angle tightness the way you index other stats. Set the league-average to 100, and any deviations from 100 represent the percent difference from the league-average (for example, 115 indicates 15% better than average). The indexing solution (in #4), suggested by Kurcon, seems simplest and, critically, a reasonable one. Using Freddie Freeman as an example, he shows that Freeman’s launch angle tightness relative to the league average was fundamentally unchanged in 2020 despite his tightness changing nominally for the worse (remember, lower is better, higher is worse): You could create a far more robust adjustment, but just comparing against league average LAstd seems to get you where you need to be. Here's Freddie Freeman on tracked pitches: LAstd (deg) / LAstd+2017: 22.1 / 1132018: 21.1 / 1162019: 21.8 / 1132020: 23.6 / 117 — Connor Kurcon (@ckurcon) November 13, 2020 So, that’s probably what I’ll do moving forward, too, until I have the time and ambition to test my proposed solution (in #3). Sorry if you thought I would update everyone with results from 2020. I will — it’s on the ol’ ideas whiteboard — but this finding from Kurcon has thrown an additional wrench in the gears. Patience, lads.