A: The VersusSportsSimulator algorithm is designed to iteratively compute team ratings based on wins and losses, with strength of schedule implicitly included in the model. These team ratings are then sorted numerically to produce the ordinal ranking of teams. However, the team ratings can only be as good as the data behind them, and if there isn’t much data then the algorithm needs some help. This is where preseason rankings come into the picture. Preseason rankings are not generated in the same way as later season rankings. In fact, they are not based on team ratings at all. They are essentially a weighted average of the final rankings of each team over the past four years. They serve as a starting point for the season and represent a view of how teams have historically stacked up against each other in recent seasons, with more recent rankings counting more than rankings from 4 years ago. Initially, as the first few regular season games are played, the preseason rankings play a large role in the computation of the weekly ratings and rankings. As the season progresses, the preseason rankings are progressively dampened out until enough win/loss data is collected to allow them to be removed altogether. VersusSportsSimulator.com completely eliminates preseason rankings from the algorithm after the 8th week of college football play. At this point, enough games have been played to objectively compute how teams rate against each other. This is also why BCS rankings are not published until after Week 7…because the computer rankings systems involved in the BCS need to have about 7 weeks of data to be truly autonomous and meaningful.

A: I think there are pros and cons to each. I like the fact that the BCS system is a hybrid between human polls and computer algorithms. This hybrid model allows human variables and observations to be factored in while still keeping a degree of objectivity. However, like most college football fans, I’d like to see the National Champion determined on the field. I don’t like the concept of two teams getting in the title game and settling it on the field while one potentially more deserving team gets overlooked. I don’t know that a 4 team playoff would completely eliminate this phenomenon, but I think the final outcome would be a lot less disputed and it would only require two additional games to be played. Since an extra week of rest would be required in the 4-team playoff model, perhaps the entire bowl season could be started a week sooner. This would somewhat alleviate the issue of teams getting rusty during the layoff between the final regular season games and the bowl games.

A: Humans just don’t have the mental capacity to simultaneously consider all of the factors and dependencies that together depict the performance of a team. Only an advanced iterative algorithm can do that. Humans tend to look at things like records and conferences to evaluate a team’s performance. Humans generally have a pretty good read on how good a team might perform against another team, based on something they’ve seen on game day or in practice, which is why human polling is equally important. I think the concept of pairing humans with computers to assess a team’s rating is the best (and only) way to get an accurate reading. My hat is off to the BCS for coming up with the formula they use. That’s not to say that I think the participants in the National Championship game should be chosen as they are today, however. I believe the BCS formula should determine the top 8 teams and the NCAA should seed those teams into a playoff. That way, the door would be open for teams that play in weaker conferences or for teams that have played well all season with just a loss or two. In the end, the best team would be named Champion, having decided it on the playing field with a 3-0 post season record.

A: Most computer generated rankings are based on some variation of wins and losses, strength of schedule, home field advantage, and margin of victory. Very few actually look at team or player level data. As this type of data becomes more and more available on the world wide web, we may see a day in which more factors are considered. But for now, I think the effort would exceed the reward. Computer algorithms are able to tell us how teams have performed against their competition and from that we can even assess the probability of an upset. But there will always be a certain margin of error (noise) that cannot be predicted by computers.

A: In 1999, Michael Vick led the Virginia Tech Hokies to the National Championship game against Florida State. The BCS was fairly new at the time and the network covering the National Championship interviewed Kenneth Massey during their coverage of the game. Kenneth and I both graduated from Virginia Tech with degrees in mathematics, and Kenneth’s undergraduate study in sports rankings earned him a spot in the computer generated aspect of the BCS formula. The concept of computer generated team rankings was new to me at the time and I was intrigued that someone so close to home could develop such a highly regarded algorithm. I began to study up on the concepts and when my computer programming job was outsourced in 2003, I decided to keep my skills sharp by developing an algorithm of my own. Kenneth Massey was instrumental in helping me formalize my ideas into something meaningful. I tweaked the algorithm over the course of several years before I finally started publishing the results in 2006. In 2007, I launched VersusSportsSimulator.com to help showcase the results and provide a public service to football fans around the country. I’ve since added several other sports to the site, including College Basketball, NFL Football, NBA Basketball, NHL Hockey, Major League Baseball, and NASCAR Sprint Cup.

A: A team’s ranking starts off during the preseason as a prediction for how that team will rank relative to all other football teams (FBS, FCS, D2, D3, NAIA, etc). The key word here is “prediction”, as no games have yet been played. These preseason rankings are based partially on a weighted average across recent past seasons, but also include a “trend” factor that indicates how that team tends to be trending (up or down) over past seasons. The preseason rankings are slowly dampened out as more and more games are played and ultimately are removed from the calculation altogether after week 7 or 8. For example, following the Week 1 games, the preseason rankings may count 90% while the actual game outcomes only count 10%. And after Week 2, the ratio might be more like 80:20. Eventually, once an adequate number of games have been played, the ratio is 0:100. The implication of ranking teams in this way is that the rankings start off as predictions, then morph into a hybrid representation of predictions and facts, and ultimately end up as “retrodictive” rankings. By “retrodictive”, I mean that they attempt to explain outcomes based on how teams fared against their competition; they shouldn't be used to predict future outcomes. That being said, the retrodictive model is based solely on wins and losses, while strength of schedule (quality of opponent) is implicitly included in the calculation.