Sunday, June 20, 2010

Let's begin

One of the great mysteries in soccer is how to grade individual performance. Standard newspaper box score is very lite on data. Scoring, booking and substitutions. That's it.


Let's try something.

Take the goals player scored during the season, add them to the team points accumulated through wins and draws plus all the positive shutouts (clean sheets). Subtract red cards and negative shutouts, cases when team lost a game without scoring a goal, and divide everything with the number of meaningful games played. That means only effective time on the pitch, not late substitutions or similar situations. Something like counting wins and saves in baseball for pitchers. If you come in to play with 3 goals ahead and your team wins a game eventually, you are not credited with a win but the player whom you relieved. It looks like this:

                            ( goals + ( wins x 3 ) + draws + shutout ) - ( red card + shutout ) / games = points

Ideally, in the league with 20 teams, maximum points would be 5. If someone played all 38 games, scored in every game and his team won all the matches and never conceded a single goal, that super player would be assigned 5 points. On the other end of the spectrum, substitute level player like, for example, Steve Finnan; good and reliable player with adequate production on averige team, would be graded around 1. Good season is around 2, great around 3 points. MVP season is around 4. Above 4 is once in a lifetime achievement.
This method strictly looks at how a player affects the bottom line: the scoreboard. It fits nicely in the 0 to 5 scale for better comprehension and is the most useful measurement available for past games. With this method we can grade Maradona's seasons or Pele's. It's not perfect, of course, but it can give us pretty good picture of the past in the absence of any other extensive source of data.

Next thing we can produce out of historical data is something similar to Win Shares in baseball. Since we're not blessed with abundance of data baseball or basketball produces, our poor man's version goes like this:

   ( ( ( wins x 3 ) + draws ) x teams points count ) / sum of all points awarded to players ( win + tie column )


                                                           In table below wins are already multiplied by 3

Witch gives us ,basically, each player's contribution for the season expressed in number of points. 3 points WS means that player was good for one straight team victory. And that is your average player, in a range between 2 to 3 points. That's Steve Finnan ( sorry Steve but those are the numbers. Better luck next year ).
In detail:


( wins x 3 ) + draws   -  For every game the player participated in, we award him a appropriate number of points ( win = 3, draw = 1, loss = 0 ) and then we sum the points for the season


x teams points count  -  Then we multiply player's points with points team gathered in the season ( for Arsenal 75 last season )


/sum of all points awarded to players ( win + tie columns )  -  Finally, we divide above number with the sum of points for all the players who got even a single point in the season ( sum of win and tie columns in table below ). The number should be around 10 times team points ( 750 )



Sometimes red card reduces number of players per match, ergo points are not divided among all 10 players. Ahem... 10. Yes, we don't count goalkeepers. They fall under different set of rules but about that in future posts.
One caveat though. Small sample size. Very rare are the players who can play all 38 games of the regular season and even then, 38 games are not enough of a number to draw anything certain out of it. But 38 is what we have. Cut off is set on 2/3 of the season i.e. 26 games. Any number below, in my opinion, is to much distorted with biases and luck.

                                                      To demonstrate;  Arsenal's season 2009/10




                                                            And some of the last season's all stars











No comments:

Post a Comment