The BCS is a funny little concoction, and I'm not the only person with thoughts on it. I am the only person whose thoughts on it are on this page, however.

New thoughts, December 2000.

For those who are not aware, the college football championship is decided in a championship game between the top two teams in the country. "Top two" is decided by a convoluted system, or rather the sum of terms calculated by several convoluted systems.

Poll rankings

There are two predominant polls, one done by the AP and the other by USAToday, in which people who are interested in the sport vote on who they think is better than whom. In the AP poll, various sportswriters list who they think are the top 25 teams in the country; the first place team is given 25 points, the second place team gets 24, and so on down the ballot. These points are then added together, so that a team that all voters agree is the best will have 25 points for each voter. The other poll works similarly, but instead of sportswriters, the votes are cast by coaches. The AP poll has a slightly better track record, but for the purposes of the BCS, a team's rankings in both of these polls are averaged, so that a team ranked number 6 by the AP and number 8 by the coaches gets 7 BCS points from the poll rankings.

Computer Rankings

Credit for computer rankings is worked out similarly; there are now eight different organizations that do computer rankings that are counted in the BCS, of which the most reputable are the New York Times and Sagarin, respectively, and each team gets points equal to the average of its 7 best rankings. If a team is ranked 4th in four of them and 6th in the other four, it gets (4+4+4+4+6+6+6)/7=4.86 BCS points for this. (The lowest ranking is excluded primarily because some of the rankings can get rather capricious; a ranking of 27th by a single computer, if it were averaged in with the seven others, would by itself introduce 3 more points than if the ranking were 3rd.)

Losses

This is easy; you get an extra point for each game you lose.

Strength of Schedule

For each team you count up the total win-loss record of all of its opponents (I think) and rank the teams by percentage. Then you divide the rank by 25, so that the team that played the best opponents gets .04 and the team that played the 63rd best opponents gets 2.52.

In some sense, the last two components are an attempt by the BCS committee itself at producing statistical rankings to add to the first two.

What I think

Some of this is relatively straightforward; the strength of schedule is the only really goofy part. The rankings averages could perhaps be improved upon by weighting the different polls/computers on the basis of previous results, i.e. how well a poll or computer has predicted games in the past; this would complicate the system somewhat and would probably not go over well with fans, to the extent that the BCS in general ever has. In any case, though, the number of computer rankings included should be pared back to a few that are well regarded.

The strength of a team's schedule should be included in some manner if the number of losses is included, but the rotisserie system is, as I said before, goofy. If all teams played very similar schedules then teams of the same strength are similarly likely to have acquired losses in the course of those games; whether the 12th schedule is really a loss better than the 37th schedule is hard to answer but is almost certainly more true some years than others. I'm not sure why they didn't simply multiply the win-loss percentage by 11, the number of games a team plays in a season, after subtracting from 1 so that lower is better as it is with the other components. This basically would give something like the RPI, and I think is considerably less goofy, not only giving more correct results but being simpler as well. The other concern I have with this is that a team could suffer for playing an extra game against a weak opponent, which would drive down the opponents' win-loss average. As I discuss elsewhere, the RPI is much less valid when a team has played a lot of opponents that aren't at about its own level of strength, and the top two teams in the country will not have found very many other teams at about their level of strength. I think, because of this, that the strength of schedule should only count the 5 best teams played, but with the proviso that any team that beat you has to be one of those 5. (A team that goes 10-1, losing to its weakest opponent, would average the top four teams and the bottom one.) In this case, you would multiply the win-loss average by 5, not 11, of course. Winning games against weak opponents doesn't mean you're good, but it doesn't mean you're bad, either, and beating five good opponents should at least be enough to throw the rest of the calculation to the polls.

More thoughts

I'm feel less strongly that the number of voices should be pared than I used to, though I do cling to the idea of weighting the votes if a good method for doing that can be found. I do feel, though, that the computer rankings are being dealt with in the wrong way. The reason the lowest ranking is thrown out is that averaging all the computer rankings gives one computer the opportunity to say that a team is no higher than thirtieth, saddle that team with an extra 3 or 4 points, and eliminate it from contention; every computer then has a veto. Now it takes two out of eight computers to veto a team. This still isn't right; neither should a poll be able to veto a team. In fact, we really shouldn't be trying to take the second best team into the championship game; we should be making sure we're taking the best team. If a lot of people and schemes think that A is better than B is better than C, and others think C is better than B is better than A, we don't need B in the game; we need A against C, because, regardless, one of those teams is the best. My current favorite idea to deal with the rankings is to average together the reciprocals of the rankings; the teams with the most first place votes, with lesser credit for second and third place votes, should play, rather than the teams that nobody thinks are thirtieth. (An undefeated team against untested teams would be likely to fit the description of a team that most people don't think is in the top two, but a significant number think is the best, and should get a chance to demonstrate it.)

I still feel largely as I did about strength of schedule. The BCS, as it's used to determine the championship game, needs to be much more adapted toward comparing the very top teams, instead of being a way of generating rankings down to the twentieth team.