LAMoneyGuy's College Football Computer Ranking

You know how the BCS uses computer ranking formulas to determine the National Championship matchup?  Well, for the non-college football fans out there, let me explain.  By the way, this has absolutely nothing to do with personal finance.  Okay then.

In the not so distant past the #1 and #2 teams in college football likely did not play at the end of the year.  Instead, prearranged bowl affiliations determined where everyone played and human polls (the Associated Press and the USA Today Coaches' Polls) determined by vote who the college football national champion was.

Well, some fellas thought this was crazy talk.  So, they created an alliance.  The alliance went through a few modifications before arriving at the current system, the Bowl Championship Series, or the BCS.  Ironically, it's not a series by any definition of the word.  Essentially, they take two human polls (the AP and the newly created Harris Poll) and combine the results with a handful of computer generated ranking systems.

Who are the guys who created these computer polls?  Nobody particularly important.  Computer geeks.  So, I decided that I would go ahead and create my own computer poll.

I won't give away any of the formulas, but I'll give you the input criteria.

• It all starts with win/loss records.  It is simple, win your games   and your odds of achieving a higher ranking will be higher.
• Strength of schedule.  How do you differentiate between undefeated   teams, one loss, two loss teams?  The primary way is a strength of   schedule factor.  A significant factor is included for your opponents   win/loss record and, to a lesser extent, their opponents win/loss record.
• Venue.  It is my belief that a home field advantage exists.  My   system rewards teams for playing road games, and in particular, tough road   games.
• Offensive/Defensive scoring component.  There is an adjustment that   rewards teams for scoring more on opponents than their opponents allow on   average, and for preventing opponents from scoring as much as they normally   score.
• Each team's opponents, and opponents' opponents venue and   Offensive/Defensive components are also a factor in the strength of schedule.
• D 1-AA games are included. Teams are better off beating good D1-AA teams   (as measured by record), but overall, their strength of schedule will suffer   for scheduling these games.
• When you win or lose is not a factor.  In other words, losing early   is not any better than losing late as seems to be the case in the human polls.    Also, if you beat a team early when they were ranked in the top 10 in the   human polls, then they descend into a 5 loss team, you get credit for beating   a 5 loss team.
• Each season happens in a void.  No win/loss factors or rankings are   brought forward from last season.
• I also don't measure yards gained or given up.  Yards don't matter,   points do.
• Head to head games do not determine which team is ranked higher.    Sometimes after a game, the losing team may remain ranked higher than the   winning team.  Each team's entire season is taken into account, and if   the winning team has an otherwise poor strength of schedule,   offensive/defensive scoring stats, even win/loss record, they will likely not   surpass the team they just beat.

I don't have any delusions of getting a phone call from the BCS.  I just thought it would be a good outlet for my inner geekhood.

Here are my results for the games through the games played on September 16, 2006.

 

                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                               

Rank	Week 3
1	Southern Cal	2.6080
2	Michigan	2.5930
3	Auburn	2.5120
4	Ohio State	2.4670
5	Louisville	2.3571
6	Florida	2.3293
7	West Virginia	2.3042
8	Rutgers	2.2845
9	Missouri	2.2517
10	Georgia	2.2007
11	Virginia Tech	2.1628
12	Alabama	2.1215
13	Boston College	2.0698
14	Boise St	2.0635
15	Oregon	2.0171
16	TCU	1.9958
17	Arizona St	1.9863
18	Iowa	1.9810
19	Michigan St	1.9432
20	UCLA	1.9427
21	Notre Dame	1.9381
22	Wisconsin	1.8081
23	Oklahoma	1.7862
24	Navy	1.7449
25	Washington St	1.7079
26	Wake Forest	1.6982
27	Maryland	1.6458
28	Texas	1.6023
29	Pittsburgh	1.5962
30	Tennessee	1.5867
31	Purdue	1.5780
32	LSU	1.5050
33	Hawai`i	1.4962
34	Kentucky	1.4956
35	Oklahoma St	1.4885
36	Iowa St	1.4770
37	South Florida	1.4556
38	Texas Tech	1.4540
39	Kansas St	1.4476
40	Indiana	1.4387
41	Clemson	1.4105
42	Nebraska	1.3952
43	Houston	1.3839
44	Ohio U.	1.3836
45	Florida St	1.3713
46	Georgia Tech	1.3683
47	Washington	1.3352
48	Arkansas	1.3309
49	California	1.2937
50	Texas A&M	1.2685
51	Minnesota	1.2600
52	Penn State	1.2073
53	Western Michigan	1.2053
54	North Carolina	1.2002
55	New Mexico	1.1804
56	Southern Miss	1.1676
57	Bowling Green	1.1368
58	South Carolina	1.1238
59	Utah	1.0914
60	Fresno St	1.0897
61	Alabama-Birmingham	1.0785
62	Mississippi	1.0607
63	Northern Illinois	1.0516
64	Arkansas St	1.0422
65	Syracuse	1.0271
66	Central Michigan	1.0194
67	Middle Tennessee St	1.0112
68	Tulsa	1.0043
69	Brigham Young	0.9924
70	UTEP	0.9903
71	Arizona	0.9737
72	Oregon St	0.9483
73	New Mexico St	0.9480
74	Northwestern	0.9461
75	Marshall	0.9400
76	Nevada	0.9265
77	East Carolina	0.9136
78	Virginia	0.8967
79	Connecticut	0.8872
80	Tulane	0.8817
81	San José St	0.8800
82	Toledo	0.8762
83	Ball St	0.8761
84	Baylor	0.8735
85	Miami FL	0.8687
86	Central Florida	0.8630
87	Wyoming	0.8276
88	Idaho	0.8143
89	Cincinnati	0.7999
90	Vanderbilt	0.7824
91	North Texas	0.7717
92	Illinois	0.7476
93	Army	0.7339
94	UNLV	0.7299
95	Kansas	0.7220
96	Colorado St	0.6850
97	Buffalo	0.6661
98	Akron	0.6659
99	Rice	0.6470
100	Louisiana-Monroe	0.6430
101	Troy	0.6372
102	Air Force	0.6342
103	Duke	0.6146
104	Kent St	0.6129
105	Stanford	0.6089
106	Florida Atlantic	0.6049
107	Louisiana Tech	0.5952
108	SMU	0.5868
109	Louisiana-Lafayette	0.5832
110	Mississippi St	0.5784
111	San Diego St	0.5687
112	Temple	0.5472
113	Eastern Michigan	0.5171
114	Memphis	0.5163
115	Florida Int'l	0.4923
116	North Carolina St	0.4382
117	Utah St	0.4191
118	Miami OH	0.4122
119	Colorado	0.376

Comments

[Charles]

When you win or lose is not a factor. In other words, losing early is not any better than losing late as seems to be the case in the human polls. Also, if you beat a team early when they were ranked in the top 10 in the human polls, then they descend into a 5 loss team, you get credit for beating a 5 loss team.

For most computer polls this seems to be the norm (i.e. quality win points for BCS). Why not modify it a bit (or do a second calculation) so that the team's W/L record when you beat them is what is held through for the rest of the season? Say two teams, #1 & #4 come into a game both 4-0. #4 wins and goes on to have an undefeated season while #1 slumps, loses three in a row and ends up 7-5. Should the (former) #4 be punished at the end of the season because they contributed to the mental and physical demoralization of former #1? College teams are rarely the same team from beginning to end - so why take into account later games?

Or... you could play with that formula even more - what about the importance of bouncing back from a loss? If you played a truly good opponent, they would win their next time around.

Just some thoughts...

[lamoneyguy]

Charles, I have played with the idea. I'm pretty handy with a spreadsheet, but I found that that additional input required an insane amount of additional formula writing.

[Charles]

I understand completely - I'd suggest using Access or some other database to play around with. The adjustments wouldn't be so bad.

[lamoneyguy]

Hey Charles,
It's also a conceptual problem. How much weighting to I assign to how good the teams appeared at the time, versus how they actually were. I'm going to tinker with that a little and see how it changes the resulte. However, I won't make any permanent changes until after the season.

Thanks for the suggestions.

[Charles]

You're welcome - let me know if you need someone to bounce ideas off of.