Quote:
Originally Posted by Rococo
I'd be curious to know how Statcast is calculating xBA. Baseball HQ has Ohtani's xBA at .336. It's a little hard for me to see how anyone with a BABIP of .391 could possibly have an xBA higher than his actual BA.
It's based on individual batted balls - it's essentially a combination of exit velo, launch angle and batter speed for each batted ball. For example certain combinations of launch angle and exit velo (and not just high exit velo) might essentially guarantee that you get a single almost all the time, in which case your xBA for that batted ball might be something like .960.
So yes, while BABIP of .391 is high, what that stat is saying is that based on Ohtani's contact quality, it should be even higher and he's been unlucky in terms where they ended up. For more info:
https://www.mlb.com/glossary/statcas...atting-average
This isn't perfect because certain hitters might persistently underperform (for example, until the shift was banned, lefty pull hitters tended to underperform xwOBA), but for a small sample size, it's far superior to traditional stats if you want to understand how the batter actually performed.
I'm sure that once there's enough data, we could improve it by just adding directional noise (as in for each batted ball, but take the average of batted balls that are within, say, 10 degrees) rather than ignoring direction entirely since batters have control over the general direction, just not precise direction. Pull tendency does matter a bit for flyball hitters with good but not great raw power since a lot of pulled HRs would become routine flyballs when hit to center. In general, you want to be a pull hitter for flyballs, but a spray hitter for groundballs.
With that said, no one is doing anything more sophisticated than Statcast as far as I know, so if someone else has drastically lower #s (well before today's game), it's either outdated, doesn't use granular data, assumes regression or maybe just wrong.
Now I think of it, adding noise to launch angle might also make expected stats more predictive as well. Batters don't control precise launch angle either but expected stats can be very sensitive to it. In fact in the future, we might even be able to simulate noise more holistically using a physical simulation (add a tiny bit of noise to contact point and timing rather than outcome).