This page attempts to explain the concepts behind the development and use of “power rankings” as a measure of form and a key feature of our prediction model.
What are Analysis FC Power Rankings?
In many cases, predictions of match results incorporate an indication of a team’s current form. This is often done looking at the win or loss ratio for the 5 or 10 most recent matches. Power rankings take this a step further by estimating of team form based on the results of recent matches as well as the quality of teams that have been faced.
One way of looking at power rankings is a simple scenario of 3 teams, A, B and C. If for example we wish to predict the result of an upcoming match between teams A and B we can use recent results against team C to provide an indication of the result. If team A defeated team C, but team B drew or lost against C then we can imply that team A will defeat team B. Though for this set of power rankings the model works on a larger scale, incorporating the full network of results across all teams in the league.
How are Analysis FC Power Rankings calculated?
Scores are calculated using a regression model where a win is scored as 100% and a loss is scored at -100%. An adjustment factor is also incorporated for home ground advantage, this normally sits at around 20%.
The final scores are between -100% and 100%. If a team wins all of its matches in the specified time it will receive a score of 100% as with that team’s involvement in a match the outcome would be perfectly correlated. If a loses all of it’s matches it will have perfect inverse correlation with the outcome and will receive a score of -100%.
Short Term and Long Term Power Rankings
We have created two sets of power rankings for inclusion in our model. They effectively follow the same methodology as mentioned above with the key difference being the time period in which they’re calculated. We have developed a short term power ranking which is based on the most recent 10 games we have data for and is useful for understanding current form. We also have power rankings which are based off the most recent 40 games we have data for which is intended to measure longer term strength or form.