Every once in a while something comes along that suggests a new window into understanding and analyzing sports. Recently in basketball it was SportVU, a set of cameras and analysis software that tracks all sorts of cool information about the players and the ball itself. Bill Barnwell at Grantland sees similar promise for that system with the NFL, but I wanted to put a little damper on things. Much like that ‘something new’ that comes along every few years, every few years there’s the suggestion that they put a chip in the football so you can tell when it’s actually broken the plane for a touchdown. The main problem with this idea, which extends to SportVU and the like, is that you still wouldn’t know if a touchdown had been scored.
The issue, of course, is that breaking the plane isn’t the only prerequisite for scoring a touchdown. A player also has to be live, or not down by contact. So if a player goes into the end zone standing up, the chip in the ball can tell you he broke the plane. But what if he gets hit right at the line and tries to extend the ball over the line? If the chip is in one end of the ball, or the middle, maybe he breaks the plane with the non-chipped end but not the whole ball. That’s a touchdown, but the chip would never know. So now you have to chip or otherwise track the entire ball. That’s a little more complicated.
Of course it gets worse. What if the player gets hit at the line but puts his hand down? That’s ok and a touchdown is scored. What if he gets hit and he puts his knee down? Or his elbow, or he spins and lands on his butt? Down by contact is a complicated rule. Now we have to have chips in or otherwise track a whole player, like with a motion-capture suit but without the ping pong balls.
Of course, it gets worse. What if a player fell without contact? What if he was in a pile at the goalline and fell on a teammate and put his knee or elbow down? The sensor would go off but he would not actually be down. We’d get a false alarm. And none of these examples even cover really contentious stuff like possession of the ball on receptions. In the end, you’re still going to have a referee or someone look at a replay and decide what happened in all but the most obvious cases.
This isn’t just nitpicking about football. You’d have similar issues with hockey, depending on the rules and the situation at the time. I imagine it would extend to other sports if I put a few minutes of thought into it. If you can’t fix the problem even by tracking every part of the ball and the players’ bodies, you don’t really have a miracle cure.
Obviously, having SportVU or motion capture or anything like that would be great. There would certainly be much more data to analyze and more ways to analyze it. But even then it would be wise to not get too ahead of ourselves. Imagine if you knew all the players on the football field at a time; I sure have. The first thing you’d do? Run an adjusted plus-minus, probably. The second thing? A regularized adjusted plus-minus. But you know what? There is absolutely no way there’s enough data in the NFL for those numbers to be meaningful. And even if there were, the selection effects would likely be overwhelming. For example, a team’s fourth receiver probably only gets on the field for pass plays; he would appear to really help passing yards while harming rushing yards. He might also only get on the field when a team is losing a lot (and hence trying to pass a lot) or for teams that successfully pass a lot (like the Packers or Patriots) and only add noise to the line-up. In contrast to the NBA, where players mostly only sub in for fatigue (some certainly do for match-ups, but I don’t think nearly as much), the NFL has lots of context-specific packages and players. And how are you going to handle special teams? Occasionally, extra data is a curse.