Programmable Umpires

Via BBTF, computers are programming human umpires:

But improve is what umpires have done every year since we’ve had access to robust PITCHf/x data. When you plot out every pitch umpires had to call a ball or a strike from Baseball Savant and judge accuracy using its strike zone definition, the consistency of this improvement stands out immediately. Over the past seven seasons, umpires have improved their accuracy by 0.37 percentage points each year, a value that seems negligible until you work out that over the course of the season the decision to call a ball or strike comes down to the umpire on roughly half of pitches thrown. This represents more than 350,000 strike zone judgement calls home plate umpires, as a group, must make each year.

This is not new information. Over the past few years, these same data have been mined to draw a similar conclusion. While the strike zone may have succumbed to gravity over the past several seasons, overall umpires are calling the true strike zone more accurately with every year.

Humans have excellent pattern recognition skills, and incentives can make them even better. Robots, rather than calling the strike zone, are training the umpires to do a better job. So by the time we build robots capable of calling the strike zone accurately, the umpires might be just as good!

There’s also a chart that shows younger umpires are better at this than older umpires, since they likely aren’t set in their ways. So progress should continue as the older umpires die out.

There is a problem with this, however. Every once in a while, MLB needs to change the strike zone to balance offense and defense. For a robot, it’s changing a constant. For a human, it’s retraining. Now, I suspect the retraining will take place faster than in the past, but it takes a while to get rid of the bias of the old strike zone.

from baseballmusings.com http://ift.tt/1W2yi0I