Friday, 16 November 2012

GW12 Predicted Scores [by SuperGrover]

Welcome to the first weekly edition of SuperGrover’s team rankings!  If you are new to the site, you can find some background on the topic here.  Before we unveil the rankings, let me walk through the process in which they are derived.

The first step in formulating team rankings is to analyze attacking and defensive performances so far this season.  Rather than use goals themselves which are prone to luck and sample size problems, I use a more complex model which combines shots on target and Big Chances to predict goals scored and conceded.  Thus, I use a team’s shots on target and Big Chances, both for and against, to determine how many goals that team should have scored and conceded.  This value is the basis of each team’s attack and defensive quality.

While valuable, those rates are not enough.  You need to adjust that rate by quality of opposition, factoring in opponent’s quality of play and home advantage.  While the impact of schedule is eventually nil, early in the season opposition strength can be badly skewed.  Hence the need to accommodate.

Determining opponent strength is done by calculating the ratio of each team’s goals scored/conceded in comparison to the average goals scored/conceded by each club on its schedule after factoring for home pitch advantage.  A clarifying example:


In this case, Team A has an attack that is 20% better than “average” and a defense that is 5% below “average” (lower defensive numbers are better).  Note that “average” is quoted as this does not represent the average of all team attack/defense ratings but rather a team who performs exactly as it’s strength of schedule would predict (i.e., 100% ratings for attack and defense).  That’s pretty much all there is to it!

Once calculated, I can use these ratings to predict scores between any two teams, home or away.  That calculation is simple; I take one team’s attacking percentage and multiply it by the other team’s defensive percentage and vice versa.  This gives me two predictive factors for each team.  I then take that factor against average goals scored in the league thus far, deriving expected goals produced for each team.  All of these factors are, of course, adjusted for home pitch advantage.

That is the process for defining my team ratings in a nut shell.  Below you will find this week’s ratings and predicted scores.  Starting next week, I will provide commentary on each week’s games, identifying fixtures in which the model simply broke down.

The Total rating below represents the combined percentage advantage over a team with attack and defensive ratings equal to 100.  It is simply a sum of attack rating over 100 and defensive rating below 100 with no additional adjustment.

As always, questions, comments and criticisms are strongly encouraged.


Projected Scores:

NB:  The score prediction model SuperGrover uses differs from the one used for the main articles, projections and captain picks on this site. 

No comments:

Post a Comment