College Football Resume Rankings - Week 1

Submitted by WestMichiganMan on

For this college football season I set out on a little project to determine what college football rankings would look like based solely on how teams performed on the field. I admit there are inherent weaknesses in this model as I have to use someone else's rankings which are not completely based on on-field results.

However here is the gist of my model: teams get awarded points for a victory, with more points being awarded for beating a higher ranked team. Rankings were from Sagarin as his were the only rankings I could find that went past the top 25 teams. Points were also given/taken away based on the point differential up to a cap of 21 points. I decided that wins/losses of 21+ were all the same, but I am considering increasing the cap. Additional points were given for away wins and taken away for home losses.

Keep in mind that I am still tinkering with the model and it is far from complete. The usual caveat of small sample sizes applies. Here is the top 15 as my model stands now:

 

Rank Team Score

1

LSU

1.000

2

Baylor

0.844

3

Boise State

0.825

4

Maryland

0.696

5

Sacramento State

0.650

6

Northwestern

0.650

7

Oklahoma

0.628

8

South Florida

0.589

9

Temple

0.563

10

BYU

0.475

11

Houston

0.443

12

Syracuse

0.380

13

Virginia Tech

0.352

14

California

0.349

15

Mississippi State

0.337

That's right, Sacramento State is the fifth best team in the country after beating Oregon State week one. The top three isn't much of a surprise and the rest of the list seems to make sense. For the record Michigan was one of a group of teams that had slightly fewer points than Mississippi State.

I plan to continue updating this model throughout the season so any and all feedback would be appreciated. I'm also considering applying this to college basketball, I think it may even be more relevant for that.

Comments

wolfman81

September 9th, 2011 at 12:05 PM ^

How is this distinct from Sagarin's rankings?  

I guess what I'm trying to ask (in a less snarky way) is how is your model going to make significant strides from Sagarin's model when you use his previous week's rankings as a starting point?  Are you planning on phasing out his rankings after this week?

Also, I'd be curious to see how this compares to other computer rankings (Sagarin, RPI, Colley...but not Billingsley).  Of course the first few weeks are really noisy and Sacremento State will probably not stay ranked quite so high (even if they keep on winning)...

 

jshclhn

September 9th, 2011 at 12:08 PM ^

Obviously Sacramento State will disappear once the sample size increases.

My biggest question is, could you expand on how your system differs from Sagarin's?

Also, what is the exact meaning behind the 1.000?  Is it like the BCS, and there's only going to be a team with a 1.000 rating if they are absolutely perfect, or is there always going to be one team with a 1.000 rating in your system?

Best of luck to you.

 

Jon06

September 9th, 2011 at 12:36 PM ^

Losses of 1-2 points should probably count for less, right? That would solve (this version of) the Sacramento State problem. Also, it's not really solely resume-based if you start with rankings, especially when the rankings that give a baseline are utterly meaningless preseason ones (if I'm right that they are).

jtmc33

September 9th, 2011 at 12:44 PM ^

They aren't the 5th best team.... they have the 5th best resume

Just think of them as that book-smart kid with the 4.0 at UM and a Yale engineering graduate degree that you would never hire at your P.R. firm  because he can't open his mouth without offending everyone arguably dumber than him and can't say a single sentence without using complex figures and engineering terminology to express his theories.

Looks great on paper, but the last man you want on the job if communicating is required.

So, if you wanted to hire a football team to beat 'Bama or OU, I'd pass on Sac State....

WestMichiganMan

September 9th, 2011 at 2:05 PM ^

To clarify, Sagarin's ratings consist of two models: one that only uses wins and losses and one that only does point differential. He has these two structured in such a way that they have a predictive element. My model isn't concerned about what will happen in the next game but only concerned with what happened in past games. Because I do incorporate points and wins to my model it is similar to Sagarin's, time will tell how similar it is. While I do incorporate his rankings into my model, this will only be for the first few weeks. I can't possibly justify awarding massive points because a team beats Sacramento State. Once the standings start to shake out, using my own rankings will become feasible.

The 1.000 is because I took each team's points and divided them by the leading teams points. This way the #1 team will always have 1.000 and the other teams' score will be their percentage of the #1 team's points i.e. Baylor has 84.4% of LSU's points.