With such a large grouping, and an efficient rating scheme (as we wrote about here), we can finally compute something resembling the first “International” ranking for Roller Derby as a whole.
An immediate objection might be “but we already have the WFTDA Rankings (and they’re making the algorithm better!)”. We agree with The Apex, that the recently released WFTDA Rankings are an improvement on previous rankings… but they still only, by design, rank WFTDA member leagues (and only their A Teams). The connected group of “rankable” teams includes many non WFTDA Members – either Women’s leagues who have not joined WFTDA yet, B Teams of members, or even Men’s leagues (who of course can’t be WFTDA members).
Another objection might be: “but FlatTrackStats already does this (and this is all done with their data!)”. We approve of everything FTS does as a service – which includes comprehensive statistical archiving, as well as their own ranking – but Flat Track Stats themselves also choose not to publish an “overall” ranking. You can see subsets of the ranking, for various sanctionings (WFTDA, MRDA, UKRDA) – and various regions (Europe, Pacific etc) – but the underlying overall ranking for all teams is not available. Additionally, FTS has the opposite problem to WFTDA: because they’re using Elo rankings, they try to rank everyone, even if a pair of teams have no competitors in common for more than a year! Our International Rankings explicitly only include the members of the largest connected set of teams in any given ranking period*, so no team can attain a rank without having played at least one bout to justify it.
This ranking, then, is based on our physical Spring optimiser, the best-in-class in our recent comprehensive review of predictive rankers. We rank the largest connected group on both score difference and (log) score ratio, and then compute a “synthetic” ranking by combining the normalised ranges of the two results. [This helps to compensate for the deficiencies of both metrics.]
So, the International Ranking (1 Oct 2016) is: (over the page, for length)