Thursday, August 2, 2012

Attempts on Goal, Shots on Goal, and a crude Expected Goal Model

Looking for a more substantial "all encompassing" metric to include in my TFS table, the decision was made to adopt an expected goal (EG) model based on the chief indicators we are provided with:  Attempts on Goal (AG) and Shots on Goal (OG). 

The first assumption made was that conversion rates are not consistent from season to season so any team displaying abnormally high or low conversion rates (both on AG and OG) would regress to the mean.  As such,  I assumed a league-wide average conversion rate for both AG and OG be applied across all teams.  Currently, that rate is approximately 10% for AG and 30% for OG.  Although OG correlates much more strongly with goals scored, for the purposes of this model I weighted OG and AG equally.  This may be subject to change going forward.

Both EG and dEG were calculated (dEG results can be found at the TFS MLS Rankings Table) and an expected goal differential (EGD) was approximated.  This was tested against the actual goal differential (EGD as X-var and GD as Y-var) and the results seem pretty favorable.  In fact, not counting the three teams with even goal differential, there was only one team with negative EGD and positive GD (DC United) and only one team with positive EGD and negative GD (New England Revolution).  Results are below.

No comments:

Post a Comment