The rankings are generated using a database of over 700,000 matches, dating as far back as 2010 for some competitions. These matches form the basis of a generalised linear model which generates an attack and defence score for each team. These scores can then be multiplied to calculate each team's ranking score.

The rankings can also be used to estimate the expected number of goals scored by each team in a match. This can be done by dividing a team's attacking score by its opponent's defensive score. If a team is playing at home, the result can then be multiplied by 1.24 to factor in home advantage.

If a team consistently scores more than is expected by the model, this will result in an increased attacking score over time, and likewise if a team concedes less goals than expected, is defensive score will improve. Thus, better than expected performance will likely result in an improved ranking for an overachieving club. Conversely, a team's scores and ranking can disimprove over time if a team consistently underperforms.

Higher importance is attached to results that happened more recently so that the rankings will reflect which teams are the strongest at a given point in time. This is achieved in the model by assigning a decay parameter to each match depending on how long ago that match took place. A match that took place a year ago is given about half the weighting of a match that took place yesterday, whereas another match that took place two years ago will have a weighting of a quarter. Therefore, matches that took place many years ago will have minimal impact on today's rankings.

This approach is the same as that used by Dixon and Coles (1997): Dixon, M. J. and Coles, S. G. (1997), Modelling Association Football Scores and Inefficiencies in the Football Betting Market. Journal of the Royal Statistical Society: Series C (Applied Statistics), 46: 265–280. In particular, I use the fbRanks R package developed by Eli Holmes (who goes into more detail about the methodology used here), with some modifications to read from a SQL database instead of CSV files.