The Andrew Extravaganza

We're in the early hours of Saturday in the fourth week of college football action; it's time to finally post something.

First up, a link to a great resource. The 2007 College Football TV Schedule, as updated on an excellent LSU fansite. I'll be using that baby to plan my Saturdays for the next three months.

Future posts will have one to three big games broken down, similar to how I did it a few years ago. I'll also be weighing in with my take on the top 25 poll. New this year is a computer power rating system that I worked up over the last week or so. I'm not sure how valuable that will be, but it's something I wanted to do for a few years now.

With three games of data to input, here is how my ratings see the college football landscape:



Eyeballing the top 25, it's look very reasonable with a few instances of screwiness, but hopefully each additional week of data will help iron things out even more.

The following block is full of skip-able information on how my ratings were derived. After that I'll throw out some betting predictions to see if my ratings can be useful for gambling. On to the explanation:

There are two components to the power ratings: team power and strength of schedule.

Team power is determined based on each team's performance in eight key statistics, weighted based on how well each particular statistic correlates to team winning percentage over the last five seasons of college football. Team power also evaluates each team's game performance. Through the magic of Excel, I smash all of those things together to get a raw team power rating.

Strength of schedule is determined by the average raw team power rating of each team's opponents, with a bonus modifier to the opponent's power rating if they were the home team. As the season progresses, all power ratings are changing, so the strength of schedule is also changing. Put simply, the strength of schedule is determined by how powerful the system views past opponents currently, it doesn't care how powerful that team was on the actual date of a game.

To get the final power rating, the raw team power rating is first normalized to a top score of 100 then modified by the team's strength of schedule, which has also been normalized to a top score of 100. The resulting value is also normalized to a top score of 100, meaning the number one team always has a power rating of 100 and every other team is rated relative to that number one team.

Over the next couple of weeks I'll be tracking the system performance to see how it does predicting game winners both straight up and against the spread. I'll probably add a link to this week's predictions sometime later today.

At this point I'll be highlighting the games which the system predicted much differently than the Las Vegas point spreads (I'll be using Caesar's Palace lines).

Stars	Away	Home	Point Spread	Prediction
	Michigan State	Notre Dame	Michigan State by 10.5	Michigan State by 36
	Arkansas State	Tennessee	Tennessee by 19.5	Arkansas State by 5
	Marshall	Cincinnati	Cincinnati by 25.5	Cincinnati by 45
	Penn State	Michigan	Penn State by 2.5	Penn State by 21
	Air Force	Brigham Young	Brigham Young by 11.5	Air Force by 7
	Ball State	Nebraska	Nebraska by 23.5	Nebraska by 6
	Purdue	Minnesota	Purdue by 14	Minnesota by 1
	Army	Boston College	Boston College by 27.5	Boston College by 42

I can see Arkansas State covering, but there's no way they win outright; Phil Fulmer would be lynched right there in Neyland Stadium, his body left hanging from the goal posts as a testament to his many failures. The Ball State at Nebraska game shouldn't be close, but if the Huskers are still reeling from the USC beat-down, I can see them struggling to bury the Cardinals. Everything else seems entirely reasonable.

Update: The Power Rating system performed pretty well, good enough to make money. Here is the complete listing for games and the system's picks for this week's slate of games.

Straight up, 33 out of 48 games were picked correctly; that's 68.8%. Against the spread, 27 out of 48 games were picked correctly, and one game was a push; that's 57.4%. The 4 Star picks went 2-1, the 3 Star picks went 1-2, and the 2 Star picks went 0-1 and had the week's only push. That means the highlighted picks went a very poor 3-4-1.

I'll be examining my prediction spreadsheet to see if I can find a trend on what the system picked well and what the system picked poorly. Based on the star selections, it might be wise to stay away from games that are way out of line with the spreads.

Turns out the trend was obvious. Games for which the system predicted score differed from the Vegas point spread by 9 points or more: 50.0% SU, 33.3% ATS. Games for which the system predicted score differed from the Vegas point spread by less than 9 points: 78.1% SU, 68.8% ATS. Very interesting, but I'll wait and see what happens in the next couple of weeks before making any conclusions.

Where's USC in the power rankings? I'm fine with them being 121, but I figure they are a little higher than that this season.

They are #6, Southern California, with a 92.0 rating.