Saturday, January 10, 2009

Who's number one?

Regardless of how you might feel about computer rankings, there's something to keep in mind when examining them. The people that come up with the models are trying to create them to be accurate. They have nothing to gain by creating a faulty model. They use some sort of input data for each team and relate it to the same type of data for every other team. Because there is some out of conference play the models can "connect" the results from those games to results of other games. Unfortunately, like the neighborhood foot race, the sample size isn't big enough for the connectedness to too meaningful. And then there's the issue of the BCS' dumb rules about not using computer rankings that have margin of victory in their model. But there's a lot of computer rankings out there that aren't constrained by those rules and it's worth looking at them.

There's a handy dandy comparison of 53 different computer rankings here.

The Gators are ranked number 1 in all but 12 of those rankings. Of those 12, the Gators took the second spot in 10. The Gators are ranked third in one and fourth in one.

The Utes took the top spot in 7 computer rankings. USC was ranked first by three of the computers. Texas was the best team according to one lonely computer. Despite their two losses Oklahoma was ranked number one by one computer.

In the computer rankings used for the BCS, the Gators were ranked second in Wolfe and Sagarin (ELO). Both had Utah number one.

When aggregated, the 53 different computer rankings give us this top five:

1. Florida
2. USC
3. Texas
4. Oklahoma
5. Utah


bryan said...

Henry, Thanks for the link to the fascinating data! I was looking at the numbers and wondered how the conferences stacked up based on the final composite computer rankings.

Some interesting things came out:

1. ACC has the highest average ranking with the SEC 2nd:
Conf Total
ACC 38
SEC 40
B12 43
BE 46
P10 51
B10 51
MWC 58
WAC 79
IA Ind 81
MAC 87
SBC 93

2. ACC also has the lowest standard deviation (16) among all conferences which lends evidence to the impression that it has the most parity.

3. Likewise, the PAC10 has the highest standard deviation (41) which demonstrates the huge gap between the top and bottom of the conference.

4. Looking at the range of rankings by taking the average +/- one standard deviation the high end of each conference would rank:
Conf High
B12 7
P10 10
SEC 13
BE 18
B10 19
MWC 19
ACC 22
IA Ind 44
WAC 47
MAC 63
SBC 74

and the low end of each conference would rank:
Conf Low
ACC 54
SEC 67
BE 73
B12 79
B10 84
P10 92
MWC 97
CUSA 108
MAC 110
SBC 111
WAC 112
IA Ind 118

All in all, these rankings seem to reinforce some common stereotypes, while revealing some surprising results (is the ACC really the strongest conference overall!?!)

Clark said...

Interesting, I think, that Utah is ranked 10th or lower by nearly 10% of the computers (5 of the 53) and has the highest standard deviation of the top 10.

Certainly, everyone that makes models or computer rankings are attempting to be accurate, but that doesn't mean they all are. Results should be reviewed the creator to make sure that they make sense, otherwise they end up publishing rubbish like:

1: USC
2: OK
3: TCU
4: FL
5: Penn State
6: Texas
7: Tulsa
8: Texas Tech
9: Boise State
10: Houston
11: Alabama
18: Utah