Arpad Elo
Using Predict
Since i'm the author of Predict, I can tell you a few things about how it works. First of all, the foundation of Predict is the ELO rating system created by Arpad Elo. You can read about the ELO system by googling 'wiki elo ratings', and you can also google 'wiki arpad elo' to read about the creator of the system. The ELO rating system is widely used in sports, online gaming, and chess to rate the skill level of competitors, and it is surprisingly accurate. Each team begins with a rating of typically 1200 or 1500, but in Predict I chose to begin each team at zero. That way teams with positive ratings are in the top half of the group (roughly) and teams with negative ratings are in the bottom half.
Then with each game the teams get points for winning and lose points for losing. The amount of points awarded or subtracted depends on the strength of the two teams. If a strong team defeats a weak team, then the change is very small. If a weak team defeats a strong team such as San Francisco defeating New England in NFL, then the change is large. Equally rated teams get a medium amount of rating change. Over time the ratings fluxuate and zig-zag their way into a pretty good estimate of how strong each team is.
The amount of points awarded or subtracted is controlled by the control variable called the k_factor, which you can see in the configuration section of the program's first screen (there are only two screens). The k_factor should be adjusted in a binary exponential sequence at first, meaning powers of two. Good values are 8, 16, 32, 64, and 128. I would use 32 as a starting point and go from there. Once you get it in the right ball park you can fine-tune the k_factor to any positive value, but I usually just leave it on one of the values I mentioned. The k_factor has the tendency to compress or expand the ratings envelope so that small k_factor values are good for sports like MLB where the teams are relatively equal but over the course of many games small differences in team strength make the difference, and large k_factor values are good for sports like NCAAFB where a strong team will crush a weak team nearly every time.