Wednesday, July 23, 2008

Baseball to Hockey, A Natural Progression

Lowetide just made a post defending the way he frequently compares hockey to baseball. First of all, it’s his blog he shouldn’t have to. His blog is top quality and the first thing I read each morning.

I've been planning a series of posts about where I think the direction the future of hockey statistics is moving but this post sparked a quick blurb here - mostly to avoid putting a 2 page comment on LT's site ;)

In the field of computer modeling, prediction, statistics (artificial intelligence) the typical development in research is to start with smaller easier to solve problems and progress to larger ones. Over the last 20 or so years there has been a progression from checkers to chess to the hottest game in AI - poker. Since 1989 researchers at the University of Alberta worked on playing perfect checkers until last year it was finally solved. Computers are able to beat humans far before achieving perfect play. In chess, a computer can beat the best human players but to my knowledge the game hasn’t been solved. Chess is a harder problem to solve than checkers because there are more possibilities and states. While in poker, pokerbots do incredibly well for themselves but the best human players can still beat the computer. This however wont last long.

I see a lot of parallels between the progression of solving games like checkers, chess and poker to the way that computers will be able to model/simulate baseball games way before being able to accurately do so in hockey games. Similar to checkers and chess, baseball is a game that has a limited set of states. When a player is batting, there is up to 3 runners on base and a limited set up outcomes. You might be able to think of 30 different events that can happen during a batting sequence but it all boils down to a limited set. For example, when a player gets beamed by a ball it’s really the same as getting walked or getting a single with no runners on base. The set of outcomes can be reduced. With hockey, the game is constantly changing, happens at such a fast pace and has tons of factors effecting every decision. In short, simulating a baseball game is much easier than a hockey game. Sports results are also different from other games because in sports, you play a team once (or even a best of seven series) and there is still a lot of variance in the expected outcome. For chess or poker, you can eliminate this by playing thousands or 10s of thousands of games. In hockey, for example, you don’t have the option of playing Sam Gagner on the powerplay 1000 times to see how he'd do. You have to make decisions based on a much smaller set of data.

As Lowetide mentioned, "baseball is wise for hockey fans because it is so much farther ahead in terms of math and math my friends is your friend when it comes to figuring things out". I think he is right and as I've tried to allude to it’s happened for a reason. As technology advances, more and more information will become available. Sites like have come up with more (and better) statistics to use when evaluating players that didn’t exist until recently. As more of this information becomes available and evolves analytical hockey fans will come up with much better ways to "suss out Bill James equations".

Keep up the good stuff LT!