Last year, we added support for a slightly modified version of the Glicko-1 rating system and I posted a pretty thorough article about it here.
I proved a few things that I already knew, but most notably that it’d be more accurate to treat split decisions and majority decisions as draws, than to treat them as full wins. Prediction rates were highest when I assigned a win value (max=1.0) of 0.55 to split decisions and 0.61 to majority decisions. This means that the loser in these cases receives the remainder of the win value. The same thing very likely applies to close unanimous decisions, but we’ll keep it simple for now. For the record, in a draw situation, 0.50 goes to both fighters.
Disqualifications are another strange outcome too. After watching them happen in MMA, gut feelings tell you that in most cases, they are freak occurrences. We have to count them in the official results, because if we didn’t, fighters would use them to bail themselves out of bad situations unless thoroughly penalized. Unfortunately, there aren’t too many of them, which makes them a poor subject for analysis.
So, what happens if we run the rating system against the pro bout database where we treat split and majority decisions as draws and ignore DQs? Well, wonder no more!