≡ Menu

Blog

The Fleming System: The Human Element of Computer Polls

Matrix tut 2

The Game Outcome Measure (GOM) is the central metric used by my rating program to determine how much better the winning team is than the losing team in a specific game played.  The GOM can be designed to reward offensively-oriented teams (by relying heavily on how many points the winning team scores); another GOM definition might favor defensively-minded teams by relying on how few points a team gives up. Still another might ignore the score altogether and just look at who wins the game.

The choice of the definition of the GOM is the one human element that enters into computer rankings.  The computer then applies the GOM criteria uniformly across all games to determine the best team.  Naturally, different GOM definitions can result in different final rankings, so it is useful to look at several criteria.

Was the margin of victory (MOV) large or small?  Was the winning team playing on the road?  Is running up the score on a weak opponent something that should be rewarded? There are many questions worth exploring when designing the GOM.  But it is important to keep in mind that the only data that can used (due to the constraints of practicality) are the final score and the location at which the game was played.  Any mathematical function that can be defined based on those data can be used.

Due to the large number of games that are played each week, it is difficult to look at any statistics other than the final score.  So, based on the score, one can make a few guesses about how a team did offensively and defensively (defined by scoring points and holding the other team from scoring points, respectively.)  Of course, the final score doesn’t tell you things like whether points were scored by offense, defense or special teams, or if a team scored a lot of touchdowns or was held to field goals.  What the final score does tell you is who won, whether it was a high scoring game or not, and if the home team defended its field or not.

There are also a number of other concerns, such as fairness and offensive bias.  Is it a good idea or a bad idea to reward a team for running up the score?  On the one hand, it is kind of a dick move, but then again, if a team trots out their second and third strings and is still scoring, it is a fair indication of the winning team’s superiority.  Offensive bias is a little bit more difficult.  Any GOM definition that includes MOV is inherently going to favor teams that are good because of their offenses.  Teams that depend on defense may be dominant, but not score many points, resulting in small margins of victory, which may be reflected in weaker GOMs.

Over the years, I have wrestled with many of these issues (as has the Bowl Championship Series). The BCS has chosen to deal with things by eliminating MOV from consideration.  This resulted in some controversy, as many computer raters felt that MOV was a necessary component to produce meaningful results.  My personal feeling is that—much like the dual-life nature of a certain feline pet of a prominent developer of quantum theory—the BCS has a point and does not have a point at the exact same time.  But much like Schrödinger’s proverbial cat, we will never really know until we move past the BCS era and see if the next system works any better.

The GOM that I currently use in my main ratings is a kind of a hybrid system, designed to reward teams for winning, while providing bonuses for MOV, defensive play, and winning on the road.  The GOM is intended to give a measure of how much better the winning team is than the losing team.  The GOM is calculated as follows:

Winning               14
MOV bonus             7 (max)
Defensive bonus       7 (max)
Defensive adjustment  -7 (max)
Road win bonus        3

There are some considerations on the bonuses.  First, the MOV bonus is capped at 14 points to avoid rewarding teams for running up the score on weaker opponents.  If the MOV is less than or equal to 14, the winning team is awarded a bonus of half of the MOV.  The defensive bonus is given for holding a team at or below 14 points.  If this is the case, the winning team is awarded a bonus of (7 –  ((losing score) / 2))  Thus, a shutout gives a team the maximum defensive bonus of 7 points.  The defensive bonus is also adjusted if the losing team holds the winning team below 14 points.  Finally, the “road win” bonus is awarded to a team that wins away from home, but is not awarded if the game is played on a “neutral” field.

All told, the maximum value for the GOM is 31 points.  The minimum is 14.5 points.  A few examples from the 2013 college football season:

Washington State - 10
Southern Cal - 7

GOM =
14      (winning)
1.5     (MOV)
3.5     (def. bonus for holding USC below 14 points)
-2      (def. adjustment for being held below 14 points)
3       (road win)
20      total

 

USC - 41
Arizona State - 62

GOM =
14      (winning)
7       (MOV)
0       (def. bonus)
0       (def. adjustment)
0       (road win
21      total

 

Notre Dame - 37
Arizona State - 34  @ Arlington, TX

GOM =            
14      (winning)
1.5     (MOV)
0       (def. bonus)
0       (def. adjustment)
0       (neutral site win)
15.5    total

 

San José State - 37
Hawai’i - 27

GOM =            
14      (winning)
5       (MOV)
0       (def. bonus)
0       (def. adjustment)
3       (road win)
22      total

Now, onto the rankings for Week 8…

The Fleming System: Week 8 Rankings

(as of October 16th)

RankSchoolWinsLossesConferenceRatingSchedule
1Florida St50ACC162.114136.914
2Clemson60ACC155.82134.57
3Missouri60SEC155.246132.746
4Alabama60SEC155.09130.006
5Oregon60PAC12153.153129.403
6Georgia42SEC152.952147.119
7LSU61SEC152.411136.054
8Ohio State60B1G150.853128.019
9Baylor50Big12150.783127.683
10Stanford51PAC12149.678135.344
11South Carolina51SEC148.513134.179
12Texas A&M51SEC148.429134.345
13UCLA50PAC12148.335126.335
14Louisville60AAC147.815122.482
15Virginia Tech61ACC147.391132.248
16Miami FL50ACC146.157123.957
17Florida42SEC146.088136.671
18Oklahoma51Big12145.393130.476
19Texas Tech60Big12144.699122.616
20Auburn51SEC144.55131.133
21Northern Illinois60MAC143.759123.425
22Washington42PAC12143.712134.712
23Central Florida41AAC143.346128.146
24Oregon St51PAC12143.122128.955
25Arizona St42PAC12143.119134.953
26Michigan St51B1G142.917127.167
27Utah42PAC12141.71135.127
28Maryland51ACC141.598127.765
29Fresno St50MWC140.391121.691
30Michigan51B1G140.12126.203
31Pittsburgh32ACC140.085136.385
32Boston College33ACC139.911140.661
33Wisconsin42B1G139.894127.978
34Southern Cal42PAC12139.606132.772
35Brigham Young42Ind-FBS139.579130.829
36Notre Dame42Ind-FBS138.362132.695
37Oklahoma St41Big12138.295123.395
38Penn State42B1G138.274130.691
39Ball St61MAC138.221123.15
40Nebraska51B1G138.139124.389
41Houston50AAC138117.4
42Mississippi33SEC137.862137.862
43Northwestern42B1G137.674131.007
44Washington St43PAC12137.411132.768
45Arizona32PAC12136.823129.823
46TCU33Big12136.698135.781
47East Carolina42CUSA136.545128.211
48Tennessee33SEC136.258135.258
49Texas42Big12135.647129.48
50Duke42ACC135.576126.992
51Iowa42B1G135.034126.118
52Tulane52CUSA134.563125.706
53North Texas33CUSA134.467132.217
54Rutgers42AAC134.177126.177
55Buffalo42MAC133.893126.06
56Ohio U.42MAC133.807125.974
57Syracuse33ACC133.765131.599
58Bowling Green52MAC133.736122.165
59Indiana33B1G133.639133.639
60Rice42CUSA133.509127.592
61Georgia Tech33ACC133.486132.57
62Boise St42MWC133.389125.139
63West Virginia33Big12133.297135.047
64Toledo33MAC132.313132.063
65Mississippi St33SEC131.908132.408
66Marshall42CUSA131.901122.568
67Vanderbilt33SEC130.998128.831
68Illinois32B1G130.623126.623
69Western Kentucky42SunBelt130.487122.07
70Colorado23PAC12130.294135.194
71Arkansas34SEC130.143133.429
72Minnesota42B1G129.662123.162
73Kansas St24Big12129.394134.394
74Navy32Ind-FBS129.202125.302
75Louisiana-Lafayette32SunBelt128.168123.568
76Wake Forest33ACC127.758126.758
77Nevada33MWC127.345127.262
78San Diego St33MWC127.179128.429
79Utah St34MWC126.651127.223
80UNLV42MWC126.251120.085
81Virginia24ACC125.952133.619
82South Alabama23SunBelt125.55128.35
83San José St33MWC125.095124.595
84North Carolina14ACC124.892137.492
85Cincinnati42AAC124.362114.028
86North Carolina St33ACC124.138126.388
87Wyoming42MWC123.571114.821
88Arkansas St33SunBelt123.469124.802
89SMU14AAC123.085137.185
90Kentucky15SEC122.744137.911
91Troy43SunBelt122.545120.045
92Iowa St14Big12122.46132.86
93Kansas23Big12122.398127.098
94Florida Atlantic25CUSA122.39129.604
95California15PAC12122.291138.708
96Texas St-San Marcos33SunBelt122.175121.842
97Louisiana-Monroe34SunBelt121.852127.71
98Kent St25MAC121.712131.64
99Memphis14AAC121.583132.383
100Middle Tennessee St34CUSA121.527126.17
101Army34Ind-FBS121.018124.947
102Central Michigan34MAC120.126124.126
103South Florida24AAC119.117129.451
104Alabama-Birmingham24CUSA118.47125.47
105Akron16MAC118.216134.358
106Texas-San Antonio25CUSA117.336126.836
107Tulsa24CUSA117125.084
108Colorado St24MWC116.975123.642
109Florida Int'l15CUSA114.143132.06
110Purdue15B1G114.079131.163
111Hawai`i06MWC112.718133.635
112Connecticut05AAC112.579133.979
113Idaho16Ind-FBS112.23130.516
114Eastern Michigan15MAC111.562127.312
115New Mexico24MWC110.894117.977
116Louisiana Tech24CUSA110.681117.598
117Temple06AAC110.542131.875
118Massachusetts15MAC109.575127.409
119Western Michigan07MAC108.279131.636
120Air Force16MWC107.784122.998
121Georgia St06SunBelt106.636129.553
122Southern Miss05CUSA105.498127.598
123New Mexico St06Ind-FBS102.191125.025
124Miami OH06MAC99.472123.222
125UTEP15CUSA98.43112.513

Patrick Fleming is a college chemistry professor at the Claremont Colleges in Southern California. He has degrees from Notre Dame and Ohio State and has done research at Tulane and Arizona State. He has been running The Fleming System since 1994. View his full catalog of rankings, including some for other collegiate sports, by clicking here.