GPI explanation

[cosmo here]

{ National Champion James Madison captured the #1 spot in the final 2004 Gridiron Power Index (GPI), the BCS-style ranking for I-AA and top indicator of at-large playoff selection. The Atlantic 10 Conference led the way with nine teams in the top 25; the Big Sky Conference and the Gateway Football Conference placed three each; the Southern Conference, the Ivy League, the Southland Conference, and the Great West Football Conference placed two teams; the Ohio Valley Conference and the Patriot League placed one each. The Atlantic 10 Conference ended with the best conference average and the GPI correctly indicated all but one playoff team correctly. Lehigh was taken as an at-large selection by the I-AA Selection Committee and the GPI indicated it would be Cal Poly. }

In looking around the I-AA.org site I found this. An explanation of the 12 components that go into the GPI would be helpful and a welcome addition to the page which would let the user know what they're looking at.

It appears that the block of text above, from the 2004 GPI page, is an apples-oranges comparison. The GPI seems to reflect the end-of-season rankings, but the Selection Committee uses the records and information available as of Nov. 21, 2004.

I would also question using 26 as a ranking for teams unranked in the top 25. It seems like this would create no meaningful difference between the 25th, 26th and 120th teams in I-AA. Maybe somebody is able to explain this away and make the numbers work.

[Tribe4SF]

Scroll to the bottom of the GPI page and there is an explanation of the components. The GPI comes out every week, so when they say what it predicted, they are talking about after the last regular season game.

Since the polls only rank 25, there's no other way to differentiate after that #.

[cosmo here]

Scroll to the bottom of the GPI page and there is an explanation of the components. The GPI comes out every week, so when they say what it predicted, they are talking about after the last regular season game.

Since the polls only rank 25, there's no other way to differentiate after that #.

thanks for pointing me to the bottom of the page. as a note, and i don't know if it changes anything, apparently the "Laz" index has changed since it was posted on I-AA.org so JMU is now first, as it should be.

it seems that a team like s car state, alabama state or coastal carolina is unfairly punished by the 26 rankings of the teams that are behind them in the three major polls. all three were ranked outside the top 40 by GPI but in most of the top 25's. is this fairly standard in the bcs and other rankings, or should they be given credit for a mention on the majority of the voters ballots.

[rokamortis]

It isn't the 26 that hurts those teams - it is the computer rankings since it takes the polls and computer rankings and then weights them.

[cosmo here]

It isn't the 26 that hurts those teams - it is the computer rankings since it takes the polls and computer rankings and then weights them.

that's understandable . . but, the human polls are still 1/3 of the final ranking. all three polls saw fit to put s car state in the top 25, but it's rendered meaningless since they're near the bottom with no real difference between them and nicholls state or cornell. so realistically, once you get one spot outside the top 25, the computer ranks everybody. there are really only four or five teams affected by this, but it still remains.

[Tribe4SF]

The computers are objective while polls are not. By the end of the season, the composite computer rankings are hard to argue with. When you're talking about teams not ranked in the polls, or ranked below #20, the computers give a pretty good view of relative strength. A team like Towson, winless in the A-10 but undefeated OOC, gets their due from the computers.

The Laz ranking tends to be out of line with the others. Some sort of algorythmic anomaly, I'm sure. (That was fun to say!) :o

[cosmo here]

The computers are objective while polls are not. By the end of the season, the composite computer rankings are hard to argue with. When you're talking about teams not ranked in the polls, or ranked below #20, the computers give a pretty good view of relative strength. A team like Towson, winless in the A-10 but undefeated OOC, gets their due from the computers.

The Laz ranking tends to be out of line with the others. Some sort of algorythmic anomaly, I'm sure. (That was fun to say!) :o

I absolutely agree that the computers are more objective, especially at this level where national exposure for most of the programs (especially on TV) is hard to find for pollsters. Not that they shouldn't give maximum effort to provide a fair ranking of the teams, but I'm more inclined to give them some slack since there's no I-AA football GameDay on the four-letter network.

The only thing I'm raising is the weighting of the human polls in the GPI. It is 1/3 for everybody, not just the top 25. At the end of the season, it seems that the polls might as well be disregarded as a measure of the top 16 automatic plus at-large berths for the tournament field.

[Lehigh Football Nation]

I absolutely agree that the computers are more objective, especially at this level where national exposure for most of the programs (especially on TV) is hard to find for pollsters. Not that they shouldn't give maximum effort to provide a fair ranking of the teams, but I'm more inclined to give them some slack since there's no I-AA football GameDay on the four-letter network.

The only thing I'm raising is the weighting of the human polls in the GPI. It is 1/3 for everybody, not just the top 25. At the end of the season, it seems that the polls might as well be disregarded as a measure of the top 16 automatic plus at-large berths for the tournament field.

The role of "human vs. computer" polls is a debate that will never be solved. Just 2 years ago fans were complaining that the BCS poll didn't have enough of a human element which allowed Oklahoma to get to the de-facto I-A championship game.

The GPI won't ever be perfect. But I think a human element with smart minds picking their top 25 is an essential part of the GPI model. And 1/3 sounds about right.