(A brief non-satirical interlude: forgive me while I indulge in a hobby topic that may bore the hell out of you. Or not.)
Another disastrous year for college football's Bowl Championship Series (BCS), where there will be, again, a disputed National Champion. The culprits appear to be the BCS computers which relegated USC (#1 in the AP Writers poll and the USAToday Coaches poll) to spectator in the "championship" game between #2 LSU and #3 Oklahoma. Well, not "computers," per se, but the mathematical algorithms used by the BCS to rank teams.
Most of the component computer ratings used by the BCS are based on some variant of the Least-Squares algorithm. There's nothing particularly wrong with Least Squares; in simple terms it is a method for finding a set of team ratings/rankings that best account for the results of games already played. The problem is that Least Squares is difficult to understand unless you have some background in regression methods or matrix algebra. As a result, a lot of college football fans have developed tinfoil hat conspiracy theories that the "computers have been rigged" to screw their favorite team.
So why not go back to human polls? For the same reason the BCS adopted computer algorithms in the first place: objectivity. But if college football fans are going to have faith in an objective mathematical ranking method, I believe it has to be open, intuitive, and something that a typical fan can compute on a hand-held calculator. Here's a modest proposal for a BCS "computer" point system.
"BLIND BALLOT" METHOD
The idea behind the Blind Ballot is simple: Each team is a poll voter, and gets to rank itself and its opponents -- and only its opponents. The more times a team wins, the more ballots it gets to cast.
1. At the beginning of the season each Division I-A team gets one ballot per Division I-A game on their schedule.
2. For each win, the team gets an additional ballot.
3. Each team ranks itself and its opponents from [1, 2, 3... (n + 1)] based on the margin of victory in their games.
4. Ballot points are awarded in the declining rank order [(n + 1)... 3, 2, 1]
5. Points are totaled across all ballots.
EXAMPLE
Here's an example using my favorite team. At the beginning of the season, Iowa had 12 D I-A games scheduled. They would start out with 12 ballots, and rank themselves #1 (one point) -- for 12 total points. Ditto Miami of Ohio: 12 ballots, 12 total points. In the first game of the season, Iowa beat Miami of Ohio 21-3; at that point Iowa would have 13 ballots, with the following rankings...
1. IOWA (2 points)
2. MIAMI - OHIO (1 point)
Miami would still have 12 ballots, with the following rankings...
1. IOWA (2 points)
2. MIAMI-OHIO (1 point)
So after the first game, Iowa would have (13 * 2) + (12 * 2) = 50 points; Miami-Ohio would have (13 * 1) + (12 * 1) = 25 points. This would be repeated over the season for each team. At the end of the regular season Iowa had 9 wins, so they would have 21 blind ballots, each having the following rankings --
1. Purdue (lost by 13) - 13 points
2. Michigan St (lost by 10) - 12 points
3. Ohio St (lost by 9) - 11 points
4. Iowa (self) - 10 points
5. Michigan (won by 3) - 9 points
6. Wisconsin (won by 6) - 8 points
7. Penn St (won by 12) - 7 points
8T. Miami OH (won by 18) - 5.5 points
8T. Minnesota (won by 18) - 5.5 points
10T. Iowa St (won by 19) - 3.5 points
10T. Arizona St (won by 19) - 3.5 points
12. Illinois (won by 31) - 2 points
13. Buffalo (won by 49) - 1 point
From Iowa's "blind" perspective Purdue is the top ranked team because they beat Iowa worse than any other team. Iowa would rank itself #4, behind each of the teams it lost to. Here, Iowa would "vote" 273 points (21 ballots * 13 points/ballot) to Purdue. Iowa would vote 210 points to itself. At the end of the regular season Iowa had 2262 total points across all ballots. It received points from 13 "voters" (its 12 opponents, and itself), an average of 174 points per voter.
I used this "Blind Ballot" point system to rank all the Division 1-A football teams based on regular season and conference championship games. Here's the top 10 based on total points...
1 Oklahoma 2933.5
2 Michigan 2748.5
3 Southern Cal 2669.0
4 Miami FL 2636.5
5 Georgia 2490.5
6 Florida St 2488.0
7 LSU 2460.0
8 Ohio St 2335.0
9 Miami OH 2322.0
10 Texas 2314.5
Here's the regular season top 10 based on points/voter...
1 Oklahoma 225.7
2 Michigan 211.4
3 Southern Cal 205.3
4 LSU 205.0
5 Miami FL 202.8
6 Georgia 191.6
7 Florida St 191.4
8 Kansas St 185.9
9 Ohio St 179.6
10 Miami OH 178.6
As a kicker, here's the updated Top 10 in total points after yesterday's bowl games...
1 Oklahoma 3219.0
2 Southern Cal 3196.0
3 Michigan 3175.5
4 Georgia 3079.5
5 Miami OH 3040.5
6 Miami FL 3024.5
7 LSU 2957.0
8 Florida St 2839.5
9 Kansas St 2699.0
10 Iowa 2656.0
... and Top 10 in points-per-voter after yesterday's bowl games...
1 Oklahoma 229.9
2 Southern Cal 228.3
3 LSU 227.5
4 Michigan 226.8
5 Miami FL 216.0
6 Kansas St 207.6
7 Georgia 205.3
8 Florida St 202.8
9 Miami OH 202.7
10 Iowa 189.7
I see a couple of advantages to the Blind Ballot method. First, it takes the mathematical mystery out of computer ratings for the layman. Anybody with a 4-function calculator can easily verify the points for a particular team -- you just need to know the outcome of the team's games and their opponent's games. Second, it is objective in the sense that every team follows the same scoring rules, and the rules reward teams that do well against tough opponents. Third, teams accumulate points as the season progresses, and the "point race" can be easily tracked.
I'd like to lobby the BCS/NCAA to adapt a method similar to this. But speaking as a human being, USC is obviously #1.
Comments