Saturday, October 24, 2009

Accounting for the Home Field Advantage

The home field advantage (HFA) is pervasive throughout sports. How to account for it when rating teams is definitely a tough nut to crack. Measuring it in college football isn't as easy as it sounds.

I mean, sure, you could just look at every D-I game, add the home and away scores, divide by games played and be done with it right? Except that non-conference games aren't scheduled randomly, and often the better teams are at home, because they can force the little guys to come to their place. If you try to figure HFA this way, you are going to overstate the impact, due to this quality leakage.

Fortunately, you can get around this by finding a random subset of games - the intra-conference schedule. From 2006-2008, in intra-conference BCS games the HFA was 2.644 points. This seems reasonable as a measurement of the magnitude of the effect. The home team won 56.7% of the intra-conference BCS games during this time-frame.

So now how do we work that into our system? We could just add 1.322 to the visiting team score, and subtract that from the home team score. That was how the WASEAN ratings were calculated in the beginning.

But reading Andrew Dolphin's work gave me a better idea. Dolphin says that when you calculate it this way, you are basically turning a win into a loss in close games, and that doesn't really make sense to do. He contends that it's more appropriate to work home field advantage into the quality of opponent adjustment. So if a team plays on the road, we treat their opponent as a better opponent. If a team plays at home we treat their opponent as a worse opponent.

This makes intuitive sense, and I've reworked the ratings based on this premise. In practice, it will cause the HFA to have more of an impact in close games than blowouts, which is a good thing. It doesn't drastically change too many of the ratings, but this seems like a better way to approach the issue, and so that's what we'll do.

The ratings tables listed in the earlier posts have been updated to reflect this change.

No comments: