After spending much time slaving over my keyboard and trying to figure out how to improve SILVA, I think I’ve finally come up with something that I can be proud of. I have now implemented version 1.1 of SILVA, a system that I feel is a very significant improvement.
While SILVA 1.0 was generally respectable at predicting fight outcomes, there were a number of ratings that were just plain wrong. Over time, I found flaws in the system that led to faulty ratings, and wondered how to go about correcting those flaws. With SILVA 1.1, I feel that I’ve done a lot to remove many of those flaws, and the result is a much improved system.
Changes have been made to just about every aspect of the system. I’ll do my best to summarize them here.
In SILVA 1.0, Victory Score had a tendency to overrate wins against opponents who had a great record against bad competition. For example, in SILVA 1.0, Jason Reinhardt received a Victory Score of 63.63 for beating 12-0 Theodore Reynol, despite Reynol having never beaten a good opponent himself. In SILVA 1.1, Victory Scores place a greater emphasis on beating opponents who have a high “strength of schedule” themselves. Now, the Reynol win counts for a Victory Score of just 49.49. Meanwhile, beating a fighter like B.J. Penn, who had a relatively low winning percentage against outstanding competition, is rated much more highly by Victory Score.
In addition, Victory Score no longer attempts to rate fights against opponents who have fewer than five career fights. This is simply because there is not enough objective information to properly rate those fights. It’s not a perfect solution, but there is no perfect solution. There’s just no way to know how good a 1-0 MMA fighter really is.
In SILVA 1.0, Fight Level was determined by taking the best win on a fighter’s record, or the average between the best win and the worst loss, whichever was lower. There were two problems created by this approach. The first was that a fighter could find himself rated very highly based on just one good win. Danny Downes, for example, achieved an above-average Fight Level just for beating Tie Quan Zhang (who was improperly rated by Victory Score in the first place). The result was that Downes was badly overrated by SILVA. The other problem was that the Fight Level of a fighter like Georges St-Pierre could be ruined by a loss to Matt Serra.
In SILVA 1.1, Fight Level looks at a fighter’s second-best win and second-worst loss. This is done to remove the chance that a fluke win or fluke loss ruins a fighter’s overall SILVA score. Keep in mind that this change was made after I recently posted the Fight Level for the featherweight and bantamweight fighters, so forget all about those.
In SILVA 1.1, rAP no longer evaluates a fighter’s entire career. Now, rAP is calculated based on a fighter’s last ten fights against experienced opponents (like Fight Level, at least five fights are needed). This is done so that a fighter like Melvin Guillard can be properly rewarded for what he’s done in his recent career, as opposed to being punished for fights he lost a long time ago.
In SILVA 1.0, SILVA was very disjointed. The fighters rated at the very top and the very bottom were rated only by rAP, while the fighters in the middle were rated only by Fight Level. In SILVA 1.1, there is one smooth formula that applies to all MMA fighters, using both Fight Level and rAP.
Additionally, I’ve taken a page from ESPN’s new Total QBR, and set SILVA on a 0-100 scale, with 50 representing an average UFC fighter. While it is still possible to have a negative SILVA, or even a SILVA of above 100, it should be much easier to interpret how good a fighter is, relative to his peers, by looking at his SILVA score.
Some comments on the new system
- While the new SILVA scores are accurate for the most part, there are some ratings that contain errors. I’ll be doing my best over the coming weeks to remove those errors. (I can tell you right now that Kyle Kingsbury will no longer be the #9 light-heavyweight.)
- I’m now providing a column giving the rank of each fighter. I was hesitant to do this in the past, because SILVA is NOT a ranking system, it’s a rating system. The idea with SILVA is to predict the future performance of fighters, not describe past performance. The other reason was simply to encourage people to look at the SILVA scores instead of the ranks. But now, I’ve included the ranks, because I feel it makes the tables just easier to look at.
- I guarantee that you will disagree with some of the ratings. I disagree with some of the ratings. If there’s one thing to keep in mind here, it’s that fighter A isn’t better than fighter B just because fighter A beat fighter B.
- Junior dos Santos is now the #1 heavyweight, and Cain Velasquez is #2. Given how much I’ve been saying that Velasquez would win their fight, I really have some mental adjustments to make.
- Jon Jones is now the #1 light-heavyweight, despite his DQ loss to Matt Hamill still being rated by the system.
- Anderson Silva has the highest SILVA score in the sport.
- SILVA no longer hates Nick Diaz. In fact, Nick Diaz is now the #2 welterweight, behind only Georges St-Pierre.
- Frank Edgar has finally received the respect he deserves, and is rated #1 at lightweight.
- Jose Aldo is now rated above Chad Mendes. I have been saying that I felt Mendes would win that fight for a long time, so again, I’ll have some mental adjusting to do.
- Inexperienced, undefeated fighters like Edson Barboza, Chris Weidman, and Paul Sass are no longer rated among the very best in SILVA. They do, however, still have high SILVA scores, just not ridiculously high.
- Declining fighters like Fedor Emelianenko, Matt Hughes, and Takanori Gomi are now rated more appropriately.
There’s so much more I want to say, but I don’t want to completely bore you to tears. Instead, check out the new SILVA scores page and take a look for yourself. Hopefully, regardless of whether or not you agree with the new ratings, you’ll find them interesting to look at.