Elo ratings are used to measure a team's relative performance to all other rated teams. After testing for the influence of variables like home field advantage (HFA) and the average goal differential between two teams of specific strengths, a poisson random number generator produces the outcome of each game. Whether a season or a complicated tournament like the Champions Leagues, the structure of the competition is modeled in a program which is then run a million times. The data produced is used to to display the results in the Sims section. The simulations are run on a regular basis throughout the season.
The Baseball and Football ELO ratings are produced by simodds.com. The NFL simulations used a normal distribution to predict number of points scored per game. This will likely be changed a bit for the 2015 season. The MLB simulations use the ELO difference between the teams to produce an outcome and the run distribution is not modeled explicitly.
The best public source for internation soccer Elo ratings is eloratings.net. However, this site uses our own calculation methodology for calculating international soccer ratings.
Elo ratings from clubelo.com are used in the all club competition simulations. In the future this will also be calculated internally.
The NFL results data comes from Pro-Football-Reference
All historical MLB results came from Retrosheet. All current season data comes from the MLB data "API".
The main problem with the entire process are the ratings. Elo is a wonderful tool that has a fair amount of predictive success. However, it is also a simplistic measure that is based only on past performance. Nothing is modeled to the level of invdividual players on teams. This means that Elo will lose some predictive power when the components of a team change. Players are transferred in and out and players get injured. Elo will only respond to these changes over time and as results dictate.