College BlogPoll Week 3: Final Ballot

The switch over to the updated RTT computer ballot thingamajig (convenient name pending) is still a work in progress, but the feedback from the rough draft gave me a lot to think over and has helped me better understand the process earlier in the season.  One of the problems I've been trying to nail down is that there's no guarantee that the metrics that correlate to total season success are effective at gauging team success early in the season.  (I.e. 12-game correlations haven't been shown to necessarily be good 3-game correlations.)  What makes the whole thing worse is that so many powerhouse programs schedule 1-AA (and equivalent) teams early on, and others play tough games (e.g. Okie State and Georgia), making the numbers very difficult to compare in any meaningful way.  With that in mind, I've been looking for better ways to weight the predictors in order to make more sense early on.  More on that after the break, but first, here's the ballot.  (Explanations are also after the break.)

Rank	Team	Delta
1	Florida
2	Alabama
3	Cincinnati	13
4	Miami (Florida)
5	Penn State
6	California
7	Arizona State
8	LSU	2
9	North Carolina	12
10	Kansas	9
11	Auburn	14
12	Iowa
13	Boise State	5
14	Texas	10
15	Oklahoma	6
16	Kentucky
17	Southern Miss
18	Houston	2
19	UCLA
20	West Virginia
21	Missouri
22	Washington
23	Clemson
24	Southern Cal	21
25	Mississippi	12
Last week's ballot

Dropped Out: Brigham Young (#7), Georgia Tech (#11), Virginia Tech (#12), Ohio State (#14), TCU (#15), Oklahoma State (#17), Utah (#18), Georgia (#22), Nebraska (#23), Michigan (#24).

Changes

• Emphasis on W/L, Strength of Schedule, Total Defense, and Total Offense.  Early on, single-game influences on the more particular metrics like passing efficiency and run defense are so heavily dependent on the opponent that those numbers get confounded.  So I de-emphasized them, which automatically increases the importance of the more generalized metrics.  I think this is better with sparse data.  Increasing strength of schedule was necessary to ensure that a slightly-better-than-average performance against a cupcake didn't outweigh an average performance against a decent team, and increasing the win-loss metric helped reflect that better teams usually do schedule cupcakes and start with winning streaks.

Exemption

• South Florida.  Having lost their starting quarterback and having only played the Little Sisters was enough to suggest that their numbers are not reliable indicators of who they are.  In other words, they are now considered an outlier.  As such, they are removed from the BlogPoll rankings this week and are probationary, pending more data.  As you see in the list, they would have been at #16.  This allowed Ole Miss to creep back in, who has only played two patsies (and even that was unimpressive) and enjoyed the late summer weekend in Oxford.

Discussion

• Cincinnati.  I suspect continued disagreement here, but the more I looked at them, the more I felt comfortable with them having a high ranking.  Numerically, they are Alabama with slightly lower defense, slightly higher offense, and a better strength of schedule.  (The defensive difference put Alabama over them).  They have a veteran quarterback in Tony Pike who currently sports a 165.0 NCAA rating (114.6 NFL rating) on 71% passing for 8 TD, 2 INT, and 923 yards.  They played Rutgers and Oregon State on the road and SE Missouri State at home, which is actually a pretty good nonconference slate this year.  Three weeks into the season, they really have had one of the best starts.  And yes, some teams do start hot and fizzle late, and that may be the case here, but it hasn't happened yet and I can't predict it.
• Kentucky and Southern Miss are two teams who have had very solid starts against poor schedules.  Unfortunately, the only way to lower teams like this is to also lower teams like Florida and Alabama - by beefing up the SOS component.  So for now, their numbers justify their placement.
• Florida State, despite having destroyed BYU, has some atrocious defensive marks.  Much of this was mediated by reducing the passing defense component (which is fair; how many teams have played two pass-wacky offenses so far?).  But bad defense is still bad defense.  Fortunately, their upcoming game against QB-deficient South Florida should remedy things and kill two birds with one stone.
• Texas and Oklahoma score virtually identically, even with the Okie loss.  This is because Oklahoma has a better strength of schedule and has actually outperformed Texas in all regards except for 3rd down offense and total offense.  Better opponents + better loss - extra loss = about the same ranking this early in the season.
• Florida, Alabama, Cincinnati, and Miami are in a class of their own right now.  They are 15-20 points above Penn State, but this is again from three games of data.

There's more work to do, and Incipient_Senescence made some very terrific points - many of which I still need to nail down.  So by all means, keep the criticism coming.  Also, I will post all data later on so you can see where all the numbers lie.