Our analysis of team and player performances in last season’s Eliteserien using expected goals (xG) and expected assists (xA) was based on Stratabet’s chance classification system where they determine the quality of a chance using six different categories, influenced by factors such as location, defensive pressure and sight of goal, with a conversion rate applied to each one.
Creating an xG model with this classification as a basis makes that it a slightly different model than the standard, but one which stands up to mathematical scrutiny (the r^2 was 0.79 for points and xG difference – if you’re into that sort of thing)
From July last season, Strata also started to include the location of each attempt on goal in Eliteserien using x and y co-ordinates. This sample of exact pitch locations for almost 3000 attempts made it possible to start building our own Eliteserien-specific xG model.
So why do this when the xG model based on Strata’s data is proven to be a good one?
Mostly because we wanted to create the first ever (we think) publicly available xG model solely based on Eliteserien data. Also, because we can and it’s (a special kind of) fun.
This Eliteserien xG model will develop and hopefully improve over time as the sample of data increases. With only about 3000 attempts, we decided to base version 1.0 of the model on only two main factors; location of the attempts splitting the pitch into ten different zones) and body part used (i.e. shot or header), in addition to penalties.
We’ve also split out attempts in Zone 6 and 7 into open passes and direct free kicks (DFKs). It was the only zones where the sample of attempts from DFKs was large enough to let us do this.
Strata’s data identifies other factors that could be incorporated into an xG model, such as defensive pressure and the amounts of opponents between the goal and the ball at the time of the attempt. We’ll look at the possibility of adding such elements into our Eliteserien xG model at a later stage but at this time the sample was just too small to incorporate them.
Below you can see the pitch location zones and the xG values we’ve calculated for each type of attempt:
The x and y locations also lets us create what we believe is the first ever Eliteserien specific xG game map. Perhaps the two best known providers of publicly available game maps are Michael Caley at (@caley_graphics) and the Twitter account @11tegen11 – our map is influenced by the work of both these excellent analysts.
For our Eliteserien map, we’ve also decided to include more descriptive data; not just if the attempt ended in a goal or not but if it was saved, missed or block and whether the attempt was a shot, header or a DFK.
(note: while in our xG game map we’ve identified each DFK regardless of location on pitch, the xG value is from the combined sample of open play and DFK attempts)
In addition to goals and expected goals tallies we’ve also added total attempts and the average xG value per attempt.
To introduce the model, we’ve used the extraordinarily Lillestrøm – Odd game last season where in addition to a hat-trick from right back Espen Ruud there was six goals scored from a combined xG value of 2.71:
Throughout the season, we’ll try to create as many xG maps as we can and use our own Eliteserien xG model to analyse player and team performances.
This article was written with the aid of StrataData, which is property of Stratagem Technologies. StrataData powers the StrataBet Sports Trading Platform, in addition to StrataBet Premium Recommendations.