Whatever you do after you read this post, do not ask me in the comments section how I got here. Because I have no clue how I stumbled on this.
That being said, I think I may have figured out a way to reduce significantly the bias/conflict of interest problems that plague the Coaches’ Poll.
To start with, I’m forced to get a little esoteric with you. I’m talking “mechanism design” economics theory here. That’s a theory that tries to bring back certain efficiencies to a market (such as insurance, where people may seek coverage without disclosing risks) where parties don’t have equal levels of knowledge about a matter. How do you find a way to share that information efficiently?
What intrigued me about that linked article was this little tidbit – “ Mechanism design has also been used to refine voting procedures in NCAA football rankings…” I googled the hell out of that fragment, but couldn’t find anything direct on it. But I did find an interview with Roger Myerson, one of the three economists who won a Nobel Prize for their work in this area, who had this to say about applying the concepts of mechanism design to voting:
… If there are 10 candidates on the ballot running, say, in a Democratic primary, you can vote for several of them, but you can’t give more than one vote to any one of them. And the person who gets the most votes, most approval votes, was approved by the largest majority, gets the bonus of winning the votes of the state.
And with that, the light bulb went off over my head. Instead of letting each coach rank in order the top 25 schools and then add up the ranked totals, why not simply have them each turn in his list of the ten best schools in the country – unranked? The teams would then be ranked in order of those which received the most votes. With 600 some odd votes for teams, you’d still be able to construct a top 25 fairly easily, I would think. (And if not, you could always let the coaches vote on a bigger group, such as a top twelve or fifteen.)
The beauty of this approach is that it minimizes, if not eliminates, the effect of any individual coach’s bias or conflict. Go back and look at Tony Barnhart’s analysis of the final regular season coaches’ poll of 2007. Remember stuff like this?
… Stoops, understandably, had his team ranked No. 1 after beating Missouri twice this season. He had three Big 12 teams in his top four (Oklahoma, Missouri, Kansas). He had LSU at No. 6.
… Somebody at Georgia made Wyoming’s Joe Glenn mad at some point. He had Georgia No. 10 on his ballot.
… You got to give the Big Ten coaches credit for one thing. They all stick together. Seven of them vote in the poll and they all had Ohio State No. 1, even Michigan’s Lloyd Carr, who has announced his retirement.
… Hal Mumme of New Mexico State continues to be our resident contrarian in the coaching fraternity. He had undefeated Hawaii No. 1 on his ballot.
With approval voting, none of those individual selections would impact the final poll, because none of the coaches would be able to vote like that.
Also, think about how the psychology of voting changes with this approach. Once a coach knows he can’t game the system with an absurdly high vote for a school (maybe to enhance the chances for a BCS crasher like Hawai’i or to improve a school’s strength of schedule by lifting other schools’ rankings), it forces him to face his vote in a more rational context. In other words, unless enough of his peers agree with his evaluation of a particular school’s merit, his vote won’t have sufficient impact. That should reduce his incentive to rig the voting.
One thing you’d have to be careful about would be to make sure that one conference wasn’t overrepresented in the voting (see Barnhart’s notes about the seven SEC coaches’ voting, and, more particularly, about the seven Big Ten coaches’ voting).
Another benefit from this approach is that I have to believe it’s much easier for a head coach to provide an honest assessment of the ten best teams in the country than to sit down and try to rank the top 25 in order. That should also increase the accuracy of the results in that they’re more likely to come directly from the coaches than from a surrogate.
I’d be interested in anyone’s thoughts on this. Have I missed something obvious?