2008 Computer Ratings

Grinnell ranked above Lebanon Valley? Bullshit.

Your system does not think well of the mighty Badgers.

Not only Arkansas State, but also Central Arkansas ranked ahead of Arkansas? That’ll leave a mark.

Records go out the window when the La Grange Rusty Forks visit the Principia Garden Gnomes Sept 27th.

Does this system take into account Principia’s 1960’s dominance of the Prarie Conference? I am not sold sir.

A&M #50? Seriously, we’re not that good yet…

That’s the best news Nebraska has heard in a while.

http://ouinsider.com/forums/showthread.php?t=73497

This guy does his own rankings and projects games based on his rankings. He’s about 75% on the season.

Any thoughts?

So the #119 team on the list put up 400+ yards on foamy-mouthed Muschamp?

Looks like a linear iterative ratings system. Like David Wilson’s except using game scores to take into account margin of victory.

Hard to say anything about them without knowing the preseason weights he’s using, etc.

012 Texas…………. 3 0 89 49 138 0.580 128.06
013 Texas Christian… 4 0 83 52 135 0.583 125.93
014 Texas Tech…….. 4 0 86 44 130 0.592 123.14
015 Tulsa…………. 3 0 92 39 131 0.587 123.04

Did the Sooner list use the alphabetical system as one of the weighted rankings?

Margin of victory is used in these rankings… The BCS would throw them out, no?

Correct. The NMV column is the ratings that don’t use margin of victory.

Nate, does the difference in his ratings equate to a neutral field line? Is that what you are calling his 75% winning percentage?

Huck, same question, would the differences in your ratings be considered a neutral field lline?

No.

The PMV column is the one that I used last year when we were comparing predictions elsewhere. Also, the ratings depress the difference relative to modern point spreads, so a multiplier should be used.

I think I used 1.25 last year but I’m not sure at all. I’ll look into what the best multiplier is based on results to date and post it on here.

Okay, so it’s actually 2.5, so double what I thought I remembered. That may be because it’s very early in the season, though. I wouldn’t consider my ratings to have stabilized yet, so I wouldn’t personally use them to analyze points spread opportunities at this point. Essentially the teams hat have shown poorly so far are rated excessively bad at this point, IMO.

But in the interest of discussion, here are the biggest discrepancies between my system’s predicted lines and the current lines (3.5 point homefield advantage plugged in – predicted margins are rounded):

Army +28.5 @ Texas A&M: Predicted margin – Texas A&M by 53.

Well, that’s not a good start. Army has been so bad that even the Aggies are a huge favorite. Right now according to the ratings Army would be a 96-point underdog at Southern Cal. I wouldn’t touch this line despite the huge confidence by the ratings. An example of what I mentioned above. I’d give the ratings at least one and probably two more weeks. The past couple of seasons there has been a 2 to 3 week window where the ratings would have cleaned up because the books hadn’t caught on to the current season’s actual team strengths yet. I think that will be in the early to mid-October timeframe.

Western Kentucky +21.5 @ Kentucky: Predicted margin – Kentucky by 46.

Same thing.

Illinois +14 @ Penn St.: Predicted margin – Penn St. by 38.

Ditto.

Northern Illinois -6.5 @ Eastern Michigan: Predicted margin – Northern Illinois by 27.

Finally, one that I might possibly bet on based on the numbers here.

Southern Cal -25.5 @ Oregon St.: Predicted margin – Southern Cal by 46.

Blowout.

UAB +24.5 @ South Carolina: Predicted margin – South Carolina by 43.

UAB is Armyesque in the ratings so far.

Navy +16 @ Wake Forest: Predicted margin – Wake Forest by 34.

Hmm.

Arkansas +28 @ Texas: Predicted margin – Texas by 45.

Hopefully.

Bowling Green -3.5 @ Wyoming: Predicted margin – Bowling Green by 20.

Is Laramie a bigger HFA than accounted for?

New Mexico -3.5 @ New Mexico St.: Predicted margin – New Mexico by 20.

It’s not that New Mexico’s good.

Troy +16.5 @ Oklahoma St.: Predicted margin – Oklahoma St. by 2.

I think this is a classic example of not enough info yet. Five of the teams’ combined six games have been against total scrubs.

There’s more, but I’m done. In a couple of weeks there will be fewer games with 10-point confidence margins, and those are the ones to play based on past performance. If anyone is interested in any particular game, let me know.

NU 5th? lol even I don’t think so

Okay, over here on the ratings post, I think I answered your question, BRAGG.

Regarding the methodology, here’s the short story. Each game is evaluated individually and each game has one total point up for grabs depending on the outcome. In the overall ratings, there is a minimum winning score that the winning team receives. The losing team receives one minus the winning team’s score, of course. In the NMV ratings, the margin of victory is ignored and each win gets the same score and each loss gets the same score. In the PMV ratings, there is no minimum winning score explicitly stated, although obviously no winning score can get as low as 1/2, although it can get very close. My ratings do not use time weighting, although I have them set up so they can if necessary, they also don’t use home or away status of the games, although I can set the program up to handle that if necessary.

Once all the games are evaluated, the program is essentially an MLE evaluator (maximum likelihood estimate). A difference in ratings between two teams is used as an estimate for the expected game score between the two teams. The difference between the actual game score and the expected game score is an error in the ratings. The probability of the ratings being correct based on the game scores is maximized.

The specific mathematics are cumbersome, and honestly, I can’t explain them very well because I don’t completely understand them myself. But the MLE evaluator that I use was developed by David Rothman and coded by Peter Wolfe. My ratings differ from Rothman’s ratings really only through the use of score ratio where I value a 20-10 win more highly than a 50-40 win because of my personal view on high scores having more variance than low scores.

I’ve also applied the ratings to various sports by tweaking the game scoring methodology to better make sense for each sport.

But at the end, no matter the sport, I take the raw ratings and evaluate each team’s sigma rating and convert to the final rating. This is something Rothman doesn’t do, but this is a better way to report, IMO. It shows essentially how many standard deviations above or below the mean each team is.

Thanks.