Tracking the ranking accuracy of the BCS computer polls.

Frequently Asked Questions

What is this web site doing?

This web site tracks the ranking accuracy of the BCS computer polls. In doing so, it exposes the worst BCS computer polls.

How are you doing that?

Each BCS computer poll publishes weekly rankings. I took the final ranking for each BCS computer poll before the bowl season began. For each bowl game I compared the rankings of both teams and used that information as a predictor of which team that poll thinks will win the bowl game. Over the course of 30+ bowl games a trend develops that describes how accurately each BCS computer poll ranks teams.

Are BCS computer polls supposed to be used like that?

Technically, the BCS computer polls are supposed to be used to rank which teams "had the best season" and are most worthy of playing in the National Championship Game.

So why are you using the BCS computer polls like that?

Because a metric like "had the best season" isn't useful at all. My attempt here is to expose the poorly performing BCS computer polls. This requires a numerical methodology. Picking the predicted winning team based on poll ranking seems logical. Moreover, it seems like the only fair methodology.

Why do you only have data for 2008-2010?

I began tracking the results only after the BCS crapfest that was 2008. If you have historical data for years before 2008, please let me know. Keep in mind I need BCS computer data which was published before the bowl games, not after.

Don't all the BCS computer polls predict the bowl winners with a high degree of accuracy?

No! For the 2008 bowl season, every BCS computer poll was below 50% accuracy -- worse than the flip of a coin!

Why don't you use the computer polls that factor in margin of victory?

Computer polls that use margin of victory are not part of the BCS rankings. The BCS rankings explicity disallow margin of victory. They're allowed to use winner-loser only. No other information about the game is allowed.

So the BCS computer polls are lousy. What do you recommend?

I have two suggestions.

- The computer polls are lousy because they're not allowed to have sufficient inputs into their algorithms. They're allowed to use winner-loser only. Thus the obvious first suggestion is to allow the BCS computer polls to use more than winner-loser. They should be allowed to use anything in a typical box score: time of possession, fumbles, interceptions, yardage, types of scores, times of scores, etc.
- Badly performing BCS computer polls should be removed and replaced by better computer polls.

Wouldn't your suggestion result in teams "running up the score?"

Certain (poor) implementations of a computer poll might. But the best implementations could detect when a team is running up the score and thus could determine that the final score does not accurately suggest the relative strengths of the teams involved. The best poll implementations would negate the potential positive effects and thereby eliminate the motivation for running up the score.

How would you replace badly performing computer polls?

Although only six computer polls may contribute to the BCS rankings, many computer polls should be tracked each season. During the offseason, the BCS committee should remove the worst-performing BCS computer poll. It should be replaced with a better-performing computer poll from the larger pool. Perhaps 100 different computer polls should be tracked each year. Relegate the worst; promote the best. This process should occur every year.

How can a computer detect a team running up the score?

It depends on the computer poll's implementation. One obvious implementation would be looking for a large disparity of points scored during the fourth quarter of a blow-out.

Wouldn't a playoff fix all this?

Yes. Yes it would.