Introduction

Like everyone else, I have opinions about the BCS. However my involvement as one of the computer programmers gives me a unique perspective and understanding of the issues that must be considered. No, I never even played high school football. But as a college football fan, I am not ignorant of the concerns of players, coaches, administrators, media, and fans.

Before 1998, bowl games extended invitations independently of each other, usually based on conference affiliations, and often well before the season was completed. It was rare for the consensus top two teams to meet in a bowl game. Often the best teams were forced to play relatively weak opponents and let the pollsters sort out the results. Split national championships, most recently 1997 Michigan and Nebraska, were regular occurences. More seriously, deserving teams (such as 1994 Penn State) were denied any share of the title. Other times, the pollsters underestimated the importance of schedule strength (1984 BYU).

Letting humans decide the national champion has many flaws, including lack of objectivity, peer pressure, inability to sort out large quantities of data, short memories, and a general lack of clear well-defined ranking criteria.

Purpose

First we define our goal of determining the most deserving team and crowning them the national champions. Of course all the subjectivity is hidden inside one's definition of "most deserving," but let's assume that we don't necessarily mean the team that would be favored by the oddsmakers.

Most people take it for granted that there is some perfect system out there that would always choose the right champion. That is simply false. Because teams don't always play to the best of their ability under ideal conditions, upsets are common in college football. The best teams in terms of strength may easily stumble during an eleven or twelve game season. Similarly, a less powerful team may benefit from lucky breaks or favorable conditions and wind up undefeated.

Therefore, it is impossible to determine the strongest team by observing a relatively short season of games where teams play vastly different schedules. The only fair criteria then for determining the merit of a team is to base it on accomplishment. A team should be rewarded for what it has accomplished on the field, not what it was supposed to have done, or what it would be predicted to do in the future.

Even an accomplishment based system is prone to dispute. In particular, how should we measure it? Even a playoff winner may have simply "gotten hot" at the right time, even though its overall season was not spectacular.

Playoffs?

Many playoff proposals have been made. It must be decided how many teams should be included, and how those teams should be selected. Logisticly, the time constraints between December and early January, and the venues of the games are also issues to consider. The most popular ideas incorporate the existing bowls, perhaps with the early rounds played at the higher seed's home field, and the later rounds at the major bowl sites. Some have proposed hybrid systems, where non-playoff teams may still be invited to minor bowls.

In reality the BCS system is a two team playoff. So the question reduces to: how many teams should make the playoffs, and how should they be selected?

In general, the fewer teams the more important the regular season becomes. College football under the current BCS format is unique in that every game matters, as even a single loss could cost a team the shot at the championship. As more teams are added to the playoff, teams may still compete for a higher seed, but without the same intense pressure and excitement. Statistically, the best team is less likely to win the tournament as the number of participants increases. Hence we routinely have non top-10 teams win the NCAA tournament. So with the goal of crowing the best team, a two team playoff may actually be best.

Selection

I believe that the playoff criteria should be well defined prior to the start of the season. Pro sports typically take the division winners and have clear guidlines for selecting wild-card teams. College sports are more ambiguous. College basketball uses a committee, which looks at factors such as polls, WL record, conference, schedule, RPI rating, and trend performances. College football's BCS formula formalizes this process for the more difficult task of picking only two teams.

For the remainder of this article, I will assume that the task is to pick two teams for a championship game. Obviously WL records are insufficient in college sports because of the number of teams and the inequity of schedules. Polls work better, but are suceptible to the human weaknesses mentioned previously. Hence we look to a formula to guide us to the best two teams.

Formulas

There has been a significant amount of research and experimentation with various computer models, and there are published methods that are optimal with respect to certain reasonable assumptions set forth before formula is derived.

Some argue that a formula takes away the human element and forces us to be tied to cold hard numbers. This is a valid concern only if we don't trust the formula. A properly designed and well understood formula is the only way to guarantee complete objectivity and the ability to set forth well defined criteria before the season starts.

BCS

Contrasting the human polls with the computers, we find many fundamental differences. It is somewhat of a paradox that although the media and public are generally ignorant of how the computer systems work, it is the computers that follow a set of predefined criteria which they implement objectively and can be duplicated independently. The human polls, on the other hand, are subject to the whims of public perception, peer pressure, and personal bias.

Some point to disagreement in the computer ratings as a sign of weakness. In contrast the AP and Coaches' polls differ very little. However, diversity is an asset, not a liability. Since the polls agree so much, there is likely an unhealthy correlation in the votes. Ballots are cast based on the "accepted" order which is infuenced heavily by media coverage and preconceived opinions. Computer ratings differ because each model looks at the same data from a slightly different perspective, independently of what everyone else is saying. By combining several computer ratings, the BCS obtains a result that may be greater than any of its parts.

Human beings are simply incapable of deciphering the complex network of games that are played over the course of a season. They have short memories, and always react in the present rather than considering the season as a whole. In particular, a late season head-to-head matchup always takes unvalidated priority. For example, in 2000 Miami was given full credit for their head-to-head win over FSU, but Washington's early season win over Miami did not receive the same emphasis. Again in 2001, Colorado got bumped up for beating Nebraska, but their early season losses to Fresno and Texas were no longer deemed important by the pollsters.

Computers can also be instructed to have decaying memories, however the BCS system requests that each game be weighed equally. Hence a team like Colorado has a difficult time digging itself out of a two loss hole. Regarding the head-to-head issue, computers don't fall into the trap of overemphasizing one game. The outcome of a game always has a variable element that may be based on many factors, from injury or weather to intangible emotions or the unfortunate bounce of the ball. Computer models incorporate an understanding of these variances and cope well with upsets. If Nebraska met Colorado again, maybe this time in Lincoln, the result might very well be reversed.

Conference championships may also be skewed by a head-to-head tiebreaker, which is a one game observation that includes a high variability. Therefore we may apply similar reasoning to argue that a conference championship should not be a prerequisite for a national championship. Nebraska and Colorado had the same conference record, but many are now arguing that the single head-to-head matchup should disqualify Nebraska while validating Colorado. Rematches often reverse the results of the first meeting, further implying that a single game is a poor yardstick when an entire season is available for analysis.

Regarding margin of victory, the media are very hypocritical. If computers incorporate MOV, then they are accused of encouraging the unsportsmanlike taboo of running up the score. Yet people lobby for Colorado because they beat Nebraska by 26 points. When computers ignore MOV, they are criticized for not recognizing how "dominant" a team has been. It is true that the score is not always a perfect reflection of the game, however it is a reasonable measure. If more complicated models (perhaps that consider the score after each quarter) are desired then they can certainly be implemented.

Perhaps the least understood aspect of computer models is the emphasis on schedule strength. Basicly, each team is measured by its performance relative to the strength of its opponents. A win over a good team gives more of a boost than a win over an undermatched team you should always beat. A consequence of interconnectivity is that each game potentially affects the rating of all the teams. TCU's win over Southern Miss should affect Nebraska's rating because it makes Nebraska's win over TCU more impressive. Apparently, computers have taught us something about the importance of schedule. In 1984 BYU won the consensus national championship with a weak schedule rivaling that of the 2001 BYU team, which never got ranked higher than 8th.

Criticism

The media and the public have submitted numerous complaints regarding the BCS system and how it has "ruined college football." Although they have the right to that opinion, it is nonsense. The BCS is an improvement over the previous system. And nobody has ever claimed it is the perfect solution. We knew from the beginning of the season that the situation that developed was possible. If there was a serious objection to the system, it should have been voiced then. All the conferences agreed to abide by the standards of the BCS. Now the teams, media, and fans may respectively disagree with the results, but cannot cry foul.

It is very easy to find fault with an imperfect system, but unless something better is proposed the criticism is worthless. Given the resistence to a playoff at this time, we must have a championship game, and therefore two teams must be chosen. I encourage college football to consider all the options and consult the necessary experts regarding this difficult task and how it relates to the public. Some controversy maintains a healthy interest and generates conversation, but this should not come at the expense of ruining the reputation of the sport.

So did the BCS get it right this year? Ultimately that depends on your definiton of "right." However, Nebraska has accomplished at least enough that we cannot label the final result as definitively wrong. The BCS did its unenviable job, and there would have been unhappy people regardless of the outcome. Congratulations to Colorado, Miami, Nebraska, Oregon, and all the other schools that should be proud of what they accomplished during the 2001 season.

Kenneth Massey
December 11, 2001