Research in Focus: Quantifying Finishing Skill
Marek Kwiatkowski, 'Quantifying Finishing Skill', StatsBomb blog
Why it's worth your time
A post from 2017, this article is proof that quality statistical work doesn't have to be in a research paper. It investigates the problem of its title while also gently explaining some drawbacks of a previous body of work, making it both research and a pretty accessible introduction to some statistical ideas.
The clarity of writing (both 'writing as laying out of thoughts' and 'writing as a sequence of words') is very worth your time if you write analytics research papers too.
What it says
Firstly, let me quote a line I highlighted: "In a low-sample sport like football, we need to wring every little bit of info from every datapoint."
This came from an argument against the practice of splitting a set of data in two to examine the repeatability of a metric, but it's a nice little philosophical challenge to analysis too.
A Bayesian inference model was used on a large sample of shots across several seasons although, given the nature of the sport, only 150 players had taken more than 200 shots. The model, like most xG models, learns how much credit various features of shots should take for resulting in goals/misses, but also attributes some credit to the players who took them.
And it finds signal even in a small amount of shots! (Albeit with very wide confidence intervals, and with a couple of other interesting notes about the method).
What's cool about it
The post's approach to lightly offering a better methodology to the question is cool, but it's made even cooler by the code being made available (you can still access it!).
From the perspective of another five years since the piece was published, it's cool looking back at the results. The model gave Robert Lewandowski a 56% probability of being above average; despite two outlandishly good scoring seasons to end his Bundesliga career, his league non-penalty goals total since 2017/18* is still only 7.8% higher than his non-penalty xG total.
*(stats from FBref, 2017/18 as far back as their xG data goes)
Finally, for a work which has an epilogue section mentioning C++ and R packages, it's a clearly football-literate piece of work. One of the drawbacks of the study that Kwiatkowski discusses is that the model "assumes single finishing skill for a player, when I can think of at least four distinct skills: stronger foot, weaker foot, head and long range". This recognition of 'finishing skill' (meaning scoring more goals than xG) as a number of separate types of finishing skill is something that isn't often explicitly acknowledged - perhaps not even recognised at all - in other work.
[Link to the article, 'Quantifying Finishing Skill', is here]
'Research in Focus' is like SparkNotes for football analytics: summarising and analysing the best research out there. Get Goalside supporters get access to posts a month early. Follow this link for the list of all Research in Focus pieces.