Sunday, 22 April 2012

Quantifying Goal Scorers.....Revisited and Expanded.

There has been some amazing work done regarding quantifying the worth of significant events that occur during a football match.The most obvious of these being goals,both scored and allowed.Much of the analysis is based around the concept of in game win and draw probability and it's derived younger brother,the Expected Points added.I include both of these concepts in my game graphs,they are also integral in determining the chances of teams progressing in various cup competitions and also the wider vista of final league position.Omar and Ford have also pioneered this approach,click on their names to read some of their posts.

The rush to provide a more context dependent explanation of goals scored can be illustrated in the following two examples.After 89 minutes of the then goalless early season game local derby between WBA and Stoke,Ryan Shotton nipped in front of a hesitant Ben Foster and prodded home a controversial winner.Stoke were looking likely to take a point from the game,but following Shotton's smash and grab raid,three points became the overwhelming favourite and the travelling Stoke fans were jubilant.

Compared this goal to Seb Larsson's 59th minute goal against The Potters a month later.It was the 4th goal in a 4-0 stroll for his Sunderland team,it merely confirmed a comfortable victory for the Wearsiders and added very little to his side's expected points total prior to his strike.Sunderland fans were no doubt happy to see another goal,but it hardly defined the afternoon,Bramble's 5th minute opener and to a lesser degree Woodgate's 11th minute own goal had already done that.

In terms of Expected Points added it is no contest.Shotton's is worth almost 2 extra points and Larsson's is worth virtually nothing.As a descriptive statistic expected points is second to none,it brilliantly capture's the importance of Shotton's goal and the relatively unimportance of Larsson's.But does it also do a disservice to the Swede.Scoring a goal against Stoke is probably about as difficult as scoring one against WBA and it's very doubtful if a player has much control over when,during a match his goal will be scored.After all,team's can only consistently improve the rate at which they score the opening goal by becoming a better side.They don't appear to be able to break out of the expected first goal rate for a team of their quality merely by trying to score first more often.Therefore,it doesn't seem fair that Larsson should be "punished" because two teammates and an errant opposing defender has managed to grab a goal or two before he has.

What's needed is a number that can be set alongside the current Expected Points Added stat to give credit to the scorer for his actual act of scoring,but is not dependent as much on the game context when he scores it.

If we divide each game into individual minute sections,there are three major game affecting things that can happen within that minute.These are,you can score a goal,you can concede a goal or much more commonly neither team will score and the importance of these events will depend on the time remaining and the current scoreline.So at the moment we are still very much in expected points territory.

If we further dissect Shotton's goal and run through all three possibilities,we find that as already noted roughly 1 expected point became 3 with his goal,but had no one scored in that minute Stoke's current EP of around one would have remained virtually the same.However,there was also the potential for Foster to sprint off his line,grab the ball before Shotton's arrival,punt the ball downfield and create a goal for WBA.Had this taken place,WBA's "goal" would have been equally leveraged and Stoke's EP of one would have crashed to almost zero.So the potential points swing,from a Stoke perspective when tied in the last minute of a game is just under 3,comprising those points they potentially receive from scoring a game breaker and those they lose through conceding at the other end.

But it doesn't end there.The overwhelmingly likely event is no goal being scored,so to find the average potential points swing for a team in the last minute of a tied game we have to include the likelihood of the two goalscoring and one no goalscoring scenarios actually occurring.If we now factor in the 90%+ chance of no one scoring we find the potential swing in this particular scenario from the perspective of one team is......very small.If we now repeat the scenario calculations for the Larsson goal we find that the potential swing in the minute that he scored his goal,leading 3-0 after 59 minutes is......even smaller still.

Our final step in the preparations to take the game situation context out of goal scoring is to calculate the potential swing for every minute of a match and for every potential scoreline differential.Hopefully,it's obvious that this figure will be larger in tied games late in the day than it will be in blowout victories also late in the day.The importance of these doable calculations lie in their use in defining an average potential swing in expected points for a football match.

We can then use this average figure as the baseline for every context specific scoreline that occurs throughout a game.The potential EP points swing for Shotton's minute in the sun at the Hawthorns,when divided by the average for all possible scenarios will generate a number in excess of one,while Larsson's minute will generate a number that will be less than one.

What we've done is describe,numerically the environment of the minute in which a player scored and it describes not only how important it would be for his team to score a goal,but also how damaging it would have been had they conceded or how neutral it would have been had they merely held station.

Shotton scored a high EP worth goal in a potentially high leverage situation....for both teams.If we divide the worth of his goal by the leverage potential of that particular minute,we will get a figure of around one.Larrson scored a very low worth EP goal in a situation of very low potential leverage,so his EP/situational EP will also be around one.In short this approach values the act of scoring and doesn't penalise players who score goals that appear unimportant,or indeed score the first goal in an epic comeback,such as the Newcastle Arsenal 4-4 from last season.It greatly reduces the effect of game situation,while retaining a flavour of the potential leverage that was present at the time of the goal.

This concept isn't new.It is used in baseball and faltering attempts have been made to incorporate it into the NFL.Hopefully I've made a decent attempt at showing how it can be used in football and why it should be used alongside the current weighted goal scoring metrics to give a fuller interpretation of a player's scoring talent where game context may inflate or reduce that estimation.

I'll stick some numbers up in due course.

No comments:

Post a Comment