Hey! Welcome to the 2004 season of my generator, whoever you are. Well, I've finally written a little program that evaluates coefficients for my graph-generated ranker, and if you measure by counting upsets, I'm afraid that it fails compared to Colley's matrix, which was able to calculate the same season with only 118 upsets. I'm still keeping it around, because of nostalgia though. And I'm moving to a system of thirds, because that's a little more accurate (by one upset) than my halving system of last season.

Order Win:     1     2            3            4            5
Score:         1     .333333333   .111111111   .037037037   .012345679

I'm also adding rankings for the NFL, both including and excluding preseason games. And in the next few weeks, as computer rankings start to become meaningful w/RT data, you might see a completely NEW ranking system join the other two.

Sooo, here are the rankings as of September 1st, 2003:

OK, I've remembered why I didn't record NFL games in the first place last year, and there are two reasons that have to do with my system:

1. First of all, NFC and AFC teams never play each other during the regular season. So what develops is a kind of bifurcation of the created graph, as the NFC and AFC sets become tangled amongst themselves without any intersecting games between them that might provide information on the relative average strength of each conference.
   This is why I'm keeping one score with the preseason in it; so that some information about interteam matchups is kept in the graph. I'm not sure how useful that will be, however. We shall have to see as the season continues.
2. Teams play each other twice in the regular season. This may be a problem, because if team A beats team B twice, my ranking system does not add any bonus points for the winning team. However, if Team A beats Team B, and is then beaten by Team B, the system treats it as both points on the board for winning and points off the board for losing. I do not know if this is a good idea or not, and will have to think about it more as the season continues.

Found out a big mistake today. I'd entered that Idaho was 1-4, with the only team they beat being Washington State. Wrong! Washington State beat Idaho on August 30, and so in their new ranking they're up at 14th place where they deserve. I've also regenerated the earlier files, if you want to go back and take a more fair look at earlier rankings.

So did I save the BCS this week? Nope! not quite!, Somehow Ohio State had enough graph-generated Karma built up from its earlier games and opponents that it is still number two in my raking system, with two losses! All USC can hope for is a number 4 slot. Maybe this will change by the end of next weekend, and the rankings will be more like what the fans want. Oh, and here's Week 12 NFL regulation rankings and Week 12 NFL with preseason included.

And the final rankings before the bowl games are...Oklahoma, then Ohio State, then Miami Ohio! Probably not what the College bowl watchers were looking for. USC comes in fifth. Some close watchers may blame Hawaii's loss to Boise state in contributing to the placement, but oddly enough, USC already had a second order win over Boise St via Oregon State, so it would not have counted either way. They were out with no chance of getting in. Here's Week 14 NFL (regulation), and Week 14 NFL (incl preseason) for those watching. Oddly enough, my system is keeping pace with Colley's in terms of postdictive accuracy, even though it wasn't really set up to handle teams playing each other multiple times.

Note to Brett Favre: It's risky to throw the ball into double coverage, but it's even riskier to throw the ball into double coverage when you have no receiver there. NFL Quarterfinals are here, with regulation season and preseason included.

I'm thinking that my method really falls apart during championship play, because the ranking scores of my teams are completely unchanged. Oddly, Colley's method doesn't seem to adjust scores as well either--Philadelphia is still ranked ahead of Carolina, even after today's game. But here are the regulation season and preseason included rankings anyway.

And that's all there is. A closer Super Bowl than anyone expected, especially Colley and I. I'm surprised at how closely my system ended up in quality to his (determined by upsets), as towards the end of the season, there was barely any change in ranking score at all--New England came out of the postseason with barely a higher score than they had coming in. Compare and contrast that to LSU's 3-point bounce in their single bowl game.

It was also interesting to see the difference in competitiveness between college and pro sports via the ranking systems. Of the 769 college games, 134 or about 1/6th ended in the lower ranked team beating the higher ranked team. In pro sports, that ratio was 69 upsets out of 267 games, or over 1 in four.

I honestly did put some effort into the 3rd ranking system I promised at the beginning of the season, but for some reason it was more unstable than I expected. It will take more work. Hopefully, I will have it ready for this autumn, and we will see how well it can rank the teams for 2005.