UPDATE 1030PM FRIDAY
MoSHPlay has some updated, but still preliminary, views of the Sunday games.
(Some of you will have noticed - and a few of you have pointed it out online - that MoSHPlay has taken a particular liking for the Dees this week. The reason for the much-higher-than-expected estimated victory margin for Melbourne comes from understanding that MoSHPlay is a MARS model that comprises an intercept and 10 hinges that trigger given different ranges of the input variables and that each have a coefficient or multiplier than amplifies the difference between the threshold for the trigger’s firing and the actual. So, for example, if one hinge is h(Expected Margin - 10) that hinge will only trigger when the Expected Margin exceed 10 points. If that hinge carries a coefficient of 5 and the Expected Margin is 20, then the contribution of this hinge to the final margin forecast will be (20 - 10) x 5 = 50 points. If, instead, the Expected Margin is 5, that hinge will not trigger.
The variables used in MoSHPlay triggers are the MoSHBODS’ expected margin for the game, and the average of the ratings of the 22 best players for each team.
Two of the hinges carry quite large coefficients and tend to fire together and be somewhat offsetting. One of those hinges - the one with the largest coefficient - has historically only fired in 0.8% of games. It’s fired this week, along with the offsetting hinge, but the offseting has been somewhat minimal. As such, we initially had the Dees by 49 but now have them by “only” 33 points.
Whether this is a case of a game with outlying and/or unusual characteristics that is, accordingly, poorly estimated by the model, or a case of a genuine quirk identified by the model, only Sunday will tell.)