Chances "On A Plate".

Goals are of course the most important statistic in a game of football. The team which scores the most goals wins the game, it's included in the Laws of the Game. Therefore, it isn't surprising that the most commonly used metric to judge attacking players is the number of goals they score.

This works fine for out and out strikers, whose value tends to be related to this bottom line, but does underrate those who are more adept at other facets of the game such as chance creation.

This has led to counting such stats as assists and chances created and including these numbers into a broad assessment of player talent.

Simple counting events does help in judging a player, but as ever raw data can mislead without context. A player who continually lays opportunities "on a plate" for his teammates, by judging the best time to release a pass so the chance may be taken without breaking stride will be judged the equal of a player who consistently over hits or misjudges the pace of a killer pass.

Looking at conversion rates for chances created by individual players may be a route to sorting the clinical from the slipshod, but this does also have major disadvantages.

A lofted precision cross to the head of an attacker who has pulled wide to the edge of the six yard box, may demonstrate great skill on behalf of the provider, but the likelihood that this perfectly executed interchange will result in a goal may be much lower than a clumsily placed pass to a teammate's feet in a closer position.

In terms of chance creation, skilled wing craft, which may be a necessary component of a diverse strategy, might consistently be under rated compared to less deft passing ability in other areas of the box.

In order to attempt to allow for the numerous shooting and headed chances that are created, it is possible to use a shot model that includes the usual ingredients, such as shot location and type, but also include the identity of the player who provided the assist.

Inevitably with limited data and an almost certainly incomplete model, some players will see the number of goals scored from their assists exceed the model's prediction, while some will fall short and it is tempting to then simply list the over achievers as being more talented at getting the ball to the eventual scorer and the under achievers as being less so.

However, this over or under performance may occur just through chance. So we need to be as sure as we can that the combination of the volume of data and the amount of deviation from the expected number of converted chances is wide enough to make talent the more likely explanation than just random variation.

Using data predominately from 2012/13 a handful of players provided passes that resulted in more goals than expected.

Give it the guy on the left!

The names are not unexpected. In my sample, Wayne Rooney was listed as the assist for 19 goals from nearly 90 chances that he created for teammates. But the model that accounts for shot type and position predicted only nine should have been scored.

However if I included the identity of Rooney as the assist in the model along with all the other chances created in the Premier League over the time frame, this "Rooney" model acknowledges that the identity of the assist is a significant component of the model.

Just as importantly, Rooney's over achieving contribution to scoring in that year's Premier League is unlikely to be down to mere chance,

By contrast, van Persie, Rooney's teammate had a much smaller over achievement when setting up chances for his teammates and in addition, this surplus was very likely just down to random variation.

In 2012-14, Ozil and Walcott were Arsenal's Rooney, with Cazorla and Ramsey their van Persie.

And while volume of chances created is obviously an important factor in team success, those who may have demonstrated an eye for the ideal pass in 2012/13 included Coutinho, Hazard, Lallana, Clyne (then both Southampton), Torres and Lukaku.

Volume without quality providers included Suarez, Mata and Silva,