Elo Predicts

AFL Some very average games this weekend in terms of skills, but some very close games also. There is no doubt that the 6-6-6 rule has had an impact on gameplay, perhaps bigger than anyone anticipated. Rating changes this week were quite small in comparison to the previous three rounds, which is an indication that team ratings across the league are a better representation of team strength. There are still only two teams with a rating of 1600 or more. If they lose to the Bulldogs (1457), Carlton's (1303) rating will drop under 1300 for the first time this season. A-League The top 6 is now set, with really only the lower positions that could change. Sydney FC (1628) are a 93% chance to finish in the top 2, while Perth Glory (1613) will get a week off. Both these teams play each other in the next round, which should be a fascinating game to watch.

AFL A better week this week, 5 correct. The expectation is that the model will correct itself, coming back to the average accuracy of around 68%. There has been a big shift in the ratings of a few teams. Let's start with the positive stories first! Brisbane leads the way with a net gain of +43, followed by Geelong (+33) and Fremantle (+29). Brisbane was predicted to miss the 8 at the start of the season, but now they are a small chance to make it. Geelong's rating tipped over 1600 for the first time in a while, and they are now ranked 1st by rating. West Coast (1603) is only just behind them but they have a net gain of only +2 this season. On the other side, the biggest story at the moment is Melbourne who has a net loss of -42 this season. This is second only behind North Melbourne (-44). The other big loser is Richmond (-32), who played one bad game and missed out on the Grand Final last year now find themselves slipping. Their current rating of 1579 is the 5th best in

AFL A very poor start to the season for the Elo model this year, which probably indicates there are some big shifts in the top and bottom teams. Coming off a season where the model was 72% accurate (and average of 70% accurate over the last seven years), initially it is hard to understand why it is performing so poorly this season. However, looking at the games in the first two rounds where the predicted winner had a 59% chance or less of winning might offer some explanation. There were 10 game in the first two rounds this season, compared with 8 last season. So the hope is that the model will be more accurate over the coming rounds. Work has started on a player-based prediction model, partly just for comparison purposes. Records for this week Biggest change: +/- 19 (Collingwood & Geelong, Richmond & Melbourne) Smallest change: +/- 2 (Port Adelaide, Carlton) A-League Perth Glory dropped a game that they were predicted to win, and the Wanderers pulled off a mass