Two stats presentations on the first day of SABR.
The first is a WPA-based look at the most exciting games of all time, based on WPA adjusted for “odds of winning the World Series.” Methodology discussion was sparse to non-existent, bulk of talk is a list of games itself. Without any detail on methodology, hard to critique or approve of the methodology, and a dry recitation of quote-unquote “exciting” games is not especially exciting in and of itself.
It’s not really clear how WPA is being used to calculate exciting games; WPA will sum to .5 for the winning team and -.5 for the losing team, or 0 for the game as a whole. I assume that he’s just taking the absolute value of WPA, but I am not sure.
He touched on a problem with this approach when he mentioned Game 7 of the 1975 World Series as being the most exciting game of that series, because Game 7 has the highest swing of probability. That’s true, but pretty much by tautology; probability will be 1 or 0 after a Game 7 of the series no matter what.
The bigger question with this approach is why. Excitement is a subjective thing, after all, and it doesn’t seem as if trying to add an objective quality to it adds anything. But even if you want to do this, I think the methodology is more interesting than the rote recitation of the list. (If you want to go methodology-light, going to just a handful of games so that you can go into detail as to why it was exciting would probably be preferable as well.) There was a list of objections, which is nice, but it’d be nicer if there was an explanation of what this approach does tell us.
The next presentation was on motion tracking to analyze baseball player mechanics. Traditional motion tracking involves wearing special suits with “markers” on them to track position – like what Andy Serkis wore when he played Gollum in Lord of the Rings. (Don’t act like you haven’t seen the special features on the Lord of the Rings DVD, either.)
But now there’s the Xbox Kinect from Microsoft, which is a cheap and widely available way to track motion of people without special marker suits, and the presenter has used an adaptation of that sensor to track pitcher motions in a training facility. The technology can’t be used in MLB, because the speed of the camera is low and so is the focal length (so you couldn’t put one in an MLB stadium without sticking it on the field).
So this company has high-speed cameras they have contracted out for, and placed telephoto lenses on them.
From here, the talk goes into the notion that the repetition of mechanics is the key to avoiding pitcher injuries. Which is something I’d like to see this kind of technology used to demonstrate, rather than simply asserted. But the system measures:
The accuracy for the system is claimed to be that of one and a half centimeters, which seems to be difficult to believe given what’s known of Pitch F/X calibration in real-world uses. (There are multiple cameras, so a 3D image is derived – this is not something that you can apply to broadcast cameras.)
Then we were shown mock-ups of a real-time display of a pitcher’s mechanics based on this data collection system. The system is currently being beta tested with the New York Mets. There are eight cameras being used to get the full 3D analysis of motion, which is well more than the three cameras currently being used in Pitch F/X. We also were shown still frames from a pitcher on the mound (although not in a live game). There will be a public proof of concept demo at SABR Analytics 2014.
There’s a lot of data involved — each pitch is 10 to 12 gigabytes of data, for terabytes of data generated per game. So analyzing the data collected is likely a massive undertaking. But it’s a potentially exciting way to look at pitching. The emphasis in the talk was on injury, but it also seems as if this could have applications in figuring out pitcher effectiveness as well.