and definitely not a strong computer programmer or statistician.

Also, I have no idea what the criteria are used in the other computer rankings, or how to properly do all the things BB’nW talks about.

But what I did do was write a program that looks at every game, and calculates the margin of victory over the course of the game, and then from the point of where the winning team gains it’s final lead on, gives basically a weighted “margin of victory”.

Basicaly if they had a 7 point lead for the final 5 minutes, they would get a score of 7*5*60 = 2100 for 7 points * 5 minutes * 60 seconds.

If they had a 3 point lead from 5 minutes left in the game until 2 minutes left in the game where they score another TD and have a 10 point lead, then their score was 3*(5-2)*60 + 10 * 2 * 60 = 1740. The loser of the game would get the negative value of the score (in this case -1740)

I figure this puts margin of victory into the game while preventing running up the score at the end, since say someone has a lead of 20 points for the last 10 minutes on, but with 1 minute left scores another TD, their score would only go from 20*10*60 = 12000 up to 20 * (10-1) * 60 + 27 * 1 * 60 = 12420 (only an increase of about 3.5%).

I also figured this could credit teams more for “dominating” a game from start to finish, (a consistent 14 point lead throughout the entire game to me is more impressive than scoring at the very end of the game to win it).

I then basically determined “weights” or “values” for each team, based on their performance through the season, that woudl be used as a multiplier on this game score to come up with a final season score. (The weights were basically adjusted by maximizing the correlation between the weight and the team’s final score when all game scores were added together).

So say you had a team that completely dominated every team it faced, was undefeated, etc, then their weight was likely adjusted to ~1, while a team that was dominated in every game had a value of ~0, and the rest of the teams were distributed throughout. This also served to basically devalue huge blowout wins against very weak teams. If you lost to a team, then the score for that game was multiplied by (1-weight) for that team (so a loss to a highly ranked team wouldn’t hurt as much since 1-0.9 = 0.1, meanwhile a loss to a weak team would hurt more since 1-0.1 = 0.9).

For final team rankings I tried a few different things with penalizing even further for losses, but I haven’t really figured out the best way to do this.

So yeah, the method isn’t completely flawless, and I’m sure there are statistical things I could do to make it better. Also, there is probably some faulty logic in the way I’m giving out scores (One such thing is that in effect a last second win or loss gives a score of ~0, which I guess makes some sense since the teams showed themselves to be about even). Maybe if I added in some sort of minimum score for a win or loss, and then this score would be added to it…I dunno.

It actually worked somewhat decently, though obviously there are some results that stick out.
Here’s where I rank the teams based on their average weighted score across all their games, multiplied by their wins divided by their losses:
1.Oklahoma
2.Florida
3.Texas
4.Boise State
5.Alabama
6.Penn State
7.USC
8.Ball State
9.Utah
10.Texas Tech
11.Missouri
12.TCU
13.Ohio State
14.Oklahoma State
15.BYU

So it appears to have some problems in ranking the BCS teams with respect to the non-BCS teams (or maybe they really are deserving of that rank…I mean for Boise State they are undefeated, and have won all their games quite decisively against not particulary very crappy teams). Also, Alabama is much lower ranked than they should be since they are undefeated (so I’m probably not penalizing losses as much as I should be, or in the way I should be…but then that would only further elevate the undefeated non-BCS teams).

I haven’t really worked on it in a bit, but it only took a couple hours of coding (most of which was spend in actually pulling the data off of the website databases and formatting/compiling/parsing it into what I needed), and I don’t really know what else I should do or want to do, since it isn’t really worth anything except as an exercise in futility (and a little bit of fun). But considering how little time effort I did put into this, and considering how much it looks as good as, or better than, many of the other computer polls out there (at least to me), I don’t think it is completely worthless.

Hey now, Pitt has excellent computer science and mathematics programs. A researcher there, Thomas Hales, in 1998 made history by verifying the Kepler conjecture. Also, they have a world-leading quantum computation laboratory.

that this effectively and completely destroyed the LGT, and a new trophy will need to be designed and made.

I still like the idea of a Lincoln Statue, top hat, and/or beard.