Can analytics help a last-placed team avoid relegation?
Let’s pretend for a moment you’re Sean Dyche, manager of Premier League bottom-dwellers Burnley FC. All that stands between you and the drop are five fixtures against beatable sides, a potential 15 points to take your team to safety. You know, grimly, that you’ll likely need a minimum of ten to survive.
In desperation to stay up, you decide to consult a stats analyst. They hand you a spreadsheet which reveals your club has the second lowest Total Shots Ratio in the league, and that you are on average conceding 15.3 shots per game whilst taking 11.6. In terms of scoring, you also know you have the second worst conversion rate in the league behind Aston Villa, but a slightly above average save percentage (all figures above are from the excellent Objective Football). You have a look at some Expected Goal (xG) tables, perhaps 21st Club’s Performance League model, and note the club is doing very poorly in expected goals for and against, yet roughly on par with their fellow relegation candidates. Nevertheless there is a silver lining—Performance League places Burnley one spot higher than their current 20th place.
So before we even consider whether any of this data is useful, let’s first agree on the basic problem—how does Burnley FC earn at least ten points in five matches against Leicester (H), West Ham (A), Hull (A), Stoke City (H) and Aston Villa (A)?
The “how” in this problem is quite open-ended. Do we know for example if there are ways Burnley can, without recourse to the transfer market, improve their current form on purpose? Or do their underlying numbers reflect the innate talent level of the squad, and are therefore immutable? This is a surprisingly little-explored area of football analytics, and yet it is presumably vital to the entire enterprise.
I am going to assume for now, based in part on how coaching changes tend to yield changes in both results and underlying numbers, that yes–clubs can intentionally improve their performance metrics, and in turn their results. While Burnley should at least accept the possibility that the deficiencies in their roster may be beyond repair via coaching, and that there are not enough games left for these changes to have a meaningful effect, at this stage they have nothing to lose in trying.
So where to begin? Generally, I would want to balance the general need for overall improvement against the particular need to take as many points as possible from five separate matches at home and away against individual opponents of varying strengths and weaknesses. In what way could the data help in coming up with a solution?
With such a poor record in conceding goals and the chance-driven nature in scoring them, perhaps the answer is to focus solely on improving defense. Yet how exactly? Through different systems? Zonal rather than man-marking? Different decisions when in possession in certain areas of the pitch? What for example worked in their 1-1 draw with Chelsea in February, beyond good fortune? A good performance analyst willing to pore through hours of game tape could work wonders here.
Also to be considered—will changing defensive habits this late in the season negatively affect Burnley’s attack? Tactically, what’s the best way to balance both? Is it a different formation? A general style of play? What situations have led to goals conceded? Set-pieces? Open-play? Counterattacks? Penalties conceded? Is there any, less obvious data we have to consider that may help here?
Additionally, is the learning curve for my squad to make these kinds of changes too steep, or are there simple, small things the team can do to vastly improve results? How would we identify these in a short period of time?
Next, I’d want to pair this general information against what we specifically know of our remaining opponents. Leicester and Hull for example are also in the bottom tier of dangerous chances conceded. Perhaps playing the counter would be the best course of action? Which players do I have at my disposal to play that kind of system? Is there a high-risk, high-payoff approach worth taking? How are either team likely to approach these fixtures?
In addition to this, I would also want to know whether in past years other clubs have performed the same feat required of Burnley, and if so how often and in what circumstances. Was there a general consensus beyond “grit and fight” as to how the team overcame the odds to stay up? Was it primarily luck? An agreeable fixture list? Or something more? A tactical shift perhaps?
Finally, I would devise a plan. Time constraints may mean coupling a general change in play with a focus on the strengths and weaknesses of individual opponents. I would re-evaluate it again and again. And I would want to keep a copy lest the board come to me after the club is relegated and tell me I didn’t do my level best to keep them up.
I’m using this as an example to illustrate several different ways to think about footballing problems. Not the abstract kind involving predictive models and all the rest, but real problems faced by clubs and managers in their daily work for which analytics per se may only provide a partial solution.
The method I outlined here is a bastardization of George Polya’s famous four step solution in his bestselling mathematics textbook How to Solve It. Whilst it’s purpose is to help students think independently in solving mathematical problems, it can also be used as a kind of DIY algorithm. The steps are:
Understand the Problem – What is the unknown? What are the data? What is the condition?
Devise a Plan – Do you know a related problem? Look at the unknown! Can you use similar problems to help find a solution?
Carry out the Plan – Check each step.
Look back – Check the result.
This column is called “Trends in Analytics,” so I will conclude by adding my own two cents in bringing up Polya’s work. The current field in football analytics is very good at many things, but not so good sometimes in identifying specific problems for which analysts may provide a partial or whole solution. Work on the latter will help further bridge the gap between analyst and club. Sometimes, it’s important for analysts to step out of R and Tableau and start to breakdown if and how clubs can actually move the needle on some of these predictive metrics. Otherwise, they are like doctors who are only able to offer a diagnosis, but not a cure.