Friday, August 15, 2008

Preseason Poll

The Sunday Morning Quarterback, utilizing his usual insight and superhuman level of college football knowledge, has assembled a top 25 preseason ranking for his own BlogPoll vote based on a detailed analysis of each college team.

We will do no such thing here.

As a statistical based analysis geek (speaking for myself), for our preseason top 25 last year I merely used a compilation of accumulated 4 years recruiting talent of college teams per Rivals.

This didn’t result in a terrible preseason poll by any means. My preseason number 1 (USC) ended up postseason number 2, and I had “national champion” LSU 6th. I had Georgia preseason 3rd (right where they ended up), Oklahoma also correct at 8th, and Tennessee and Auburn among those within a few places of their final spot. Errors included Florida (I had them 2nd), Michigan (5th for me), and ranking FSU and Miami in the top 10 (ugh).

This season I am going to take a slightly different, though entirely statistical based, tact. Instead of using accumulated talent, I have created a equation that combines the average talent ranking of the 110 teams for which I have data combined with their win-loss percentage from the previous season. In this way I hope to capture not only the talent level, but the recent effectiveness with which that talent has played.

USC is number one in my poll using this method. They get there as follows –

4.01 (the average “star” rating of USC players per Rivals) + 0.846 (USC’s win percentage last season) = 4.856 (USC’s total score)

Introducing my top 25 (with totals) –

1 Southern Cal 4.856
2 LSU 4.555
3 Ohio State 4.519
4 Florida 4.440
5 Georgia 4.439
6 Texas 4.404
7 Oklahoma 4.376
8 Michigan 4.227
9 Tennessee 4.067
10 Florida State 4.046
11 Auburn 3.980
12 Clemson 3.972
13 Alabama 3.948
14 Penn State 3.897
15 Miami-FL 3.847
16 Notre Dame 3.785
17 Virginia Tech 3.763
18 California 3.721
19 Oregon 3.717
20 UCLA 3.717
21 Missouri 3.695
22 Kansas 3.683
23 Arizona State 3.679
24 South Carolina 3.660
25 Nebraska 3.639

How accurate will this prove to be? I have no earthly idea, but perhaps no more inaccurate than any other preseason effort. In looking at the results, I don’t particularly like FSU at 10th. I do think Florida coming in .001 better than Georgia is about perfect, however.

What is notable about this is how teams, at least conceptually, can be compared. While the aforementioned Florida and Georgia are for all purposes equal, USC is 6.6% “better” than LSU by total score.

So there it is. Perfect? By no means.

But better than anything else I could hope to guess my way to.

And since the work is done, the other teams by ranking are-

26 Boston College 3.616
27 Texas A&M 3.601
28 Arkansas 3.558
29 Illinois 3.557
30 Virginia 3.530
31 Texas Tech 3.530
32 Wisconsin 3.525
33 West Virginia 3.521
34 Oklahoma State 3.506
35 Maryland 3.484
36 Arizona 3.459
37 Pittsburgh 3.372
38 Iowa 3.353
39 Oregon State 3.310
40 Georgia Tech 3.291
41 Louisville 3.290
42 Michigan State 3.288
43 Colorado 3.264
44 Purdue 3.250
45 North Carolina 3.233
46 Mississippi State 3.218
47 Brigham Young 3.199
48 Rutgers 3.195
49 Kansas State 3.167
50 Hawaii 3.143
51 Ole Miss 3.143
52 Utah 3.137
53 South Florida 3.135
54 NC State 3.117
55 TCU 3.085
56 Washington 3.080
57 UCF 3.057
58 Wake Forest 3.050
59 Boise State 3.044
60 Fresno State 2.990
61 Kentucky 2.988
62 Stanford 2.981
63 Northwestern 2.945
64 Southern Miss. 2.921
65 Connecticut 2.897
66 Cincinnati 2.894
67 Washington St. 2.844
68 Troy State 2.834
69 Tulsa 2.827
70 New Mexico 2.775
71 Indiana 2.736
72 East Carolina 2.725
73 Houston 2.718
74 Minnesota 2.716
75 Memphis 2.711
76 Iowa State 2.703
77 Vanderbilt 2.677
78 Syracuse 2.649
79 San Diego St. 2.631
80 Baylor 2.620
81 Toledo 2.602
82 Louisiana Tech 2.569
83 Wyoming 2.562
84 Miami-OH 2.532
85 Ohio 2.515
86 Middle Tenn St. 2.507
87 Nevada 2.499
88 Western Michigan 2.477
89 San Jose St. 2.474
90 Louisiana-Mon. 2.468
91 Marshall 2.428
92 Air Force 2.412
93 Duke 2.393
94 UAB 2.379
95 Temple 2.366
96 Eastern Michigan 2.353
97 Colorado State 2.340
98 Tulane 2.336
99 Kent 2.335
100 UNLV 2.327
101 Rice 2.325
102 Louisiana-Laf. 2.320
103 New Mexico St. 2.273
104 North Texas 2.267
105 SMU 2.251
106 Navy 2.243
107 Northern Illinois 2.224
108 Utah State 2.179
109 Idaho 2.131
110 Army 1.910

15 comments:

Anonymous said...

Never mind FSU 10th. Michigan 8th?

I just don't get the point to preseason polls.

Mergz said...

Yeah, I should have pointed out Michigan too. Attempting to adjust to a new offense is going to be tough.

In the end I wanted to create something relatively objective, as any attempt to be subjective at this point is pure folly. In fact, as you point out, preseason polls are themselves folly.

Anonymous said...

Just because recruits are based on the 1-5 star rating, your formula says that recruiting is 5 times more important than getting W's (is this the z0oOo0k formula?)

What if recruiting services gave players from 1 to 10 stars? Would you still simply add up both components? Tres arbitrary, no?

In your formula, a team that pulled in all 4-star recruits and lost EVERY game last year would be ranked ahead of a team that pulled in all 2-star recruits and ran the table.

Meh!

Henry Louis Gomez said...

First of all any weighting of the two values would be arbitrary, so that's kind of a non-issue. Secondly the recruiting element would only be worth 5 times the win % if it was 5 and the win % was 1.000 or some other combination. In the case of a 1-star school with a .5000 winning percentage the recruiting element would count as twice as important not five times as you assert. So it's variable.

Lasty, in Mergz' defense I'll add that he begins with the premise that preseason polls are horseshit but within the horseshit his had some overlap with reality at the end of last season. In other words this is a tweak. Giving win% from last year the same theoretical weight as the recruiting prowess of the last four years would be more than a tweak.

The foundation is that the teams should be ranked according to the amount of talent they have accumulated recently (according to "experts") and adjusted slightly for how well they have been able to use that talent recently.

Mergz said...

What Henry said.

Mergz said...

Additionally, I essentially throw this poll out after the first game is played.

Anonymous said...

OMG, did you get your math education from F$U???

Star rating gets you from 1-5 points.
Win percentage gets you from 0-1 point.

Even if you recruit from the nearest homeless shelter (you and I are both 1-star recruits according to Rivals), you get as many points (in this system) as you do for a PERFECT season. Since your POTENTIAL gain from stars is 5 times as much as from winning percentage, don't fall into the trap that for those with mediocre recruits, it's "variable."

Wighting is "arbitrary?" and so it's a non-issue? That's my point - the original numbers are arbitrary (Rivals could have gone from 1-10 stars, making this system even LESS respecting of winning percentage). If arbitrary numbers makes things a non-issue, you just trivialized everything you've ever written!

Henry Louis Gomez said...

FIrst of all (I'm trying really hard to stay civil) sir, I just showed you an example in which the the recruiting number is only worth twice as much as the win %. It works in reverse too. If you have a 3 star team with a .500 record then the recruiting component is worth SIX times the win % from the previous year, not the FIVE that you state. It's variable whether you want to believe it or not.

As for arbitrary, the whole F'ing exercise is arbitrary. You're nickname is very appropriate. Duhh.

Anonymous said...

Did you mean "slightly different ... tack" instead of "tact"?

Jams said...

Would it not make sense to multiply winning percentage by average stars, rather than adding?

It's clearly a multiplicative relationship by nature anyway, not an additive one. Average "star rating" tells how potentially good a team is; the winning percentage tells how successful the coaching staff was at bringing out that potential.

It's clearly not going to be perfect (nor will any preseason poll be) but it certainly seems more logical than adding the arbitrary numbers.

Henry Louis Gomez said...

There's certainly nothing about that suggestion that makes it less valid. But it's still arbitrary. Any combination will be by necessity. I'll reiterate that last year Mergz only looked at recruiting. And he was somewhat pleased that some of the picks ended up being pretty good. The idea here was to tweak it a little. To penalize teams that did poorly with good talent and reward teams that did well with inferior talent. Multiplying does really amplify the winning percentage portion of the equation. I suspect that Mergz would feel that it's too much.

Jams said...

True, it probably places undue emphasis on last year's win percentage. I was thinking maybe
(s)*(1+p), where s is the number of stars and p is the win percentage. But I suppose that is getting a little unnecessarily complicated for a ballot that will just be disregarded after the first week.

Out of curiosity, where was the data for this culled from?

Mergz said...

James -

The data on the "stars" came from Rivals. It is an average of the talent for the past 4 years.

I'm curious as to what (s)*(1+p) might show, so I'll run it and see

Mergz said...

Using James' formula we get -

1 Southern Cal
2 LSU
3 Ohio State
4 Georgia
5 Texas
6 Oklahoma
7 Florida
8 Michigan
9 Tennessee
10 Auburn
11 Clemson
12 Penn State
13 Florida State
14 Virginia Tech
15 Kansas
16 Missouri
17 Alabama
18 Arizona State
19 Oregon
20 Boston College
21 West Virginia
22 California
23 Miami-FL
24 Illinois
25 Virginia

Not a whole lot of change at the top, but it does knock Notre Dame out of the top 25, a good thing probably.

Jams said...

Yeah, that's pretty interesting. Not a whole lot of difference, but thanks for running it.

Seems like the hardest thing to put into stats is teams that are steadily improving (or getting worse).

I guess we shall see what happens this season.