AFLW Round 5 Results - Some Regression
/WoSHBODS tipped 6 from 9 winners this week, and recorded a Margin MAE of 18.7 points per game and a Totals MAE of 18.2 points per game. That took it to a season long 76% accuracy, 18.2 Margin MAE, and 21.1 Totals MAE.
We can review WoSHBODS’ Accuracy and MAE by subdividing games based on the final margin, which we do in the table at right.
It shows that WoSHBODS is now 30 from 32 in games that have been won by 2 goals or more, but also that its MAE is under 14 points per game across all the games won by under 4 goals.
In games won by less than 2 goals, WoSHBODS is only 4 from 13 in terms of accuracy, which is worse than chance.
We can also look at how WoSHBODS has performed on a team-by-team basis, firstly in terms of MAE, which we do in the table at left.
It shows that the final margins in games involving Hawthorn, Essendon, Sydney, or Fremantle have been very well forecast - generally within just over two goals of the actual margin - while those in games involving Gold Coast or West Coast have been less well forecast and in error by as much as 4 or 5 goals.
Overall, game margins have been forecast with an MAE of around 3 goals, which feels entirely acceptable.
We can also review how well WoSHBODS has estimated the victory probabilities of each team by looking at the log probability score recorded in those games in which they were involved, which we do in the table at right.
It reveals that WoSHBODS has done well at estimating win probabilities for games involving Melbourne, Adelaide, GWS, or Geelong, and less well at games involving Fremantle, Collingwood, Brisbane, or St Kilda.
For those four teams, LPSs have actually been negative.
On wagering, a small loss was recorded by the Combined Portfolio after Gold Coast failed by a few points to protect its 30.5 points start. Overall now, the Combined Portfolio is up by 2.9c, that from a +8% ROI on a 0.38 turn.
Overs/Unders betting aside, which accounts for only 5% of the Combined Portfolio, we can review the wagering performance of the two main Funds by making separate calculations for home versus away status, and favouritism versus underdog status, which we do in the table at left.
It reveals that the most lucrative wagering in terms of ROI has come from line wagering on away team favourites, head-to-head wagering on home team underdogs, and line wagering on away team underdogs.
Least lucrative has been line wagering on home team favourites, and head-to-head wagering on away team underdogs.
TEAM DASHBOARD
In the latest Ranking on Dashboard Metrics chart, the metric rankings currently most highly correlated with the competition ladder rankings are:
MoS Win Production Function: +0.99
Points Conceded and Goals Conceded: +0.93
Scoring Shots Conceded: +0.91
Own Points Scored: +0.87
Own Goals Scored: +0.86
Q4 Performances: +0.84
% of Quarters Won: +0.81
Own Scoring Shots: +0.78
The metric least correlated is:
Q3 Perfomances: +0.34
Opponent Scoring Shots: +0.35
And, lastly, below is the full extended version of the Team Dashboard.
It shows, among other things, that no team has won more than 17 of the 20 quarters they’ve played in, and that GWS has won only three. It also shows that Adelaide have generated 10.6 more scoring shots than their opponents across entire games, including 6.6 of them in second halfs.
Also, Collingwood have scored 42% of their points in Q1s, but only 6% in Q4s, while also conceding 45% of the points that they have in Q4s but only 9% in Q1s.
In a similar vein, Carlton have conceded 42% of the points that they have in Q4s, and Fremantle 44% in Q3s. Melbourne have conceded only 3% of the points that they have in Q4s. GWS have scored 41% of their points in Q3s but only 1% in Q4s, while Western Bulldogs have scored 41% of their points in Q3s but only 8% in Q4s.
It’s hard not to think that conditioning isn’t playing a large part in the outcomes for some teams.
GWS are still yet to kick a goal in Q4s, whilst Melbourne has yet to concede one in Q4s.
