The GPI, Modified GPI and Major Polls
by, 10-25-2011 at 11:55 PM (1228 Views)
A quick look at the GPI and the two major polls, the FCS Coaches Poll and The Sports Network Poll displays some notable differences.
In fact, the two polls contain the exact same teams save one different one in each.
20% of the same teams in both polls are not in the GPI top 25 (Alabama State, Delaware, Indiana State, Jackson State, Tennessee Tech and William & Mary) and the FCS Coaches Poll has an additional one (North Dakota) which is not in the GPI top 25. Illinois State is in the Sports Network top 25 and in the GPI, but not in the Coaches.
So while the three rankings are about 80% the same teams (we'll leave out the specific ranking placement here), there is significant variance.
As you can see below, there are 44 rankings compared weekly at http://masseyratings.com/cf/compare1aa.htm - the website of Kenneth Massey (a ranking specialist and advisor to the GPI and BCS):
2. Laz Index*#
10. Massey MOV*
16. Round Robin Win %
21. Sport Theory
22. Massey BCS#
23. Dolphin Pred
24. Random Walker FL
25. Dokter Entropy
26. Ashburn BCSC#
28. Sagarin Predictor
36. Any Given Saturday
37. Real Time RPI
39. The Rank
40. FCS Now
42. Xpected Win %
43. FCS Coaches Poll*#
44. Sports Network*#
* - included in the GPI
# - included in the system used by the NCAA FCS Playoff Committee
The Gridiron Power Index has been listed first for the FCS/I-AA every week it has existed.
This is because the GPI includes the top computer ratings despite including the polls which as shown above, contain teams the vast majority of systems rank outside of the top 25.
For instance, Jackson State is ranked in the top 20 in the polls but #66 by the GPI. A look at the voters in the polls reveals the reason for the obvious bias.
The main difference between the regular GPI and the modified GPI which the NCAA FCS Playoff Selection Committee uses for teams in leagues that qualify but do not have a automatic qualifying bid is the regular version uses a mix of different computer rankings, the modified version contains none that use margin of victory in their formula and only ones that weight all NCAA games the same, and use strength of schedule. For instance the Sagarin ranking does not weight FBS, FCS, and non-Division I teams with the same formula and so is not included in the modified version.
Also, the regular GPI uses seven computer rankings, removing the high and low rank for each team to produce an average number (ARC).
The final GPI number uses the ARC and the two polls to produce a ranking number, with the five computers and two polls being equal.
The modified GPI uses the five computer rankings as an ARC ranking and the two polls as additional numbers (three numbers in all) which the NCAA considers equal.
Thus the computers are weighted differently in the two manifestations but if taken as a component of a single number, more heavily in the regular GPI (71.4%) than in the modified GPI (33.3%).
Finally it should be noted that no major poll has done better than the GPI at indicating the eventual at-large playoff team selections and no one from CSN votes in any of the major polls since being excluded by the Sports Network in 2010.