After some superb discussion at the World Conference on Science and Soccer folks back in 2014 I first wrote this article; I’ve since improved the Total Soccer Index (though I haven’t published the new algorithm).
Here’s my most recent article on predictability (Predicting Team Standings in Professional Soccer) – it was this article that led to me presenting the PWP TSI Predictability topic at the WCSS 2017 (Rennes, France) last year.
Anyhow… back to this article.
To do this we agreed that Goals Scored needed to be removed from the equation. In doing that here’s how the Composite PWP Index offers up an Expected Wins Index number for MLS teams (both home and away) and then separately for home and away games.
I took this extra step since most feel or think that teams who play at home have a better chance of winning than teams who play away from home; and indeed the average for home teams this year is 1.58 points, per game, versus away teams being 1.09 points per game.
Caveats prior to the diagrams:
- There will always be an issue with the sample size when looking at a predictability model based upon team performance (within a single game) as opposed to specific individual actions by players like shots taken and shots on goal – repeatability here is not those specific instances of activity but the ‘comprehensive instances’ of ‘team activities within a game’…
- For me that represents the primary bell curve of the game as opposed to, in my view, the 3rd or 4th standard deviations of the bell curve where goals are scored… I hope that makes sense???
- There are only 17 home and 17 away games – again that sample size is extremely small – but the volume of team activities within those games ‘should’ provide a general (bell curve) picture on what regular activities a team takes in order to score goals and/or prevent goals being scored.
- The Predictability Indices are the PWP Indices (minus) the percentages of Goals Scored versus Shots on Goal.
- No additional analysis to go with the CPWP, APWP and DPWP Predictability Indices – in my view there simply aren’t enough data points yet; when all the teams have reached 10 games played (both at home and away) I’ll break open the analysis as a real predictive tool to try and predict a win, draw or loss.
- For now consider they represent a ‘visual diagram’ that also includes the current Correlation (R) the Indices have with respect to total points or an average of points per game played in MLS.
The Composite PWP Predictability Index (PI): The CPWP PI has a best Correlation with respect to Average Points earned in the MLS League Table (R) .52
The CPWP PI for Home games has a best Correlation with respect to Average Points earned in the MLS League Table (R) .66:
The DPWP PI for Home games has a best Correlation with respect to Average Points earned in the MLS League Table (R) .60:
The APWP PI for Away games has a best Correlation with respect to Sum of Points earned in the MLS League Table (R) .58:
The visual diagram with the best Correlation to Points in the MLS League Table is CPWP-PI with an R of .66; and the second best is the DPWP-PI with an R of -.60.
So from a, point going forward approach, it would seem to me that the best visual diagram to use when offering up analysis later this year is the CPWP-PI compared with “Average” Point taken per game as opposed to the “Sum” of Points in the MLS League Table.
We will see if that holds true in a few weeks time; thanks for your patience.