I couldn’t resist making the title of this post ‘the moonwalking bear’. Many of you will have already seen the video I linked above, I think I saw it twice in separate presentations at both the Sports Analytics Innovation Summit and the Science + Football Conference earlier this year, but the point it makes is useful to repeat from time to time.
Being the pedant that I am, I’d like to mention that the bear hardly moonwalks – it’s more like an awkward jig through the teams – but that argument is for another day and another place where I can air my opinions on moonwalking bears to my heart’s content.
And now I’ll come clean, this post isn’t actually about a moonwalking bear. Time to click the ‘back’ button on your browser if you’re expecting to see more people dancing whilst dressed up as animals. A tiger doing the macarena perhaps? Sorry.
So this post is all about what we don’t see, what we don’t record, and what we don’t tend to look for.
The Numbers Game summarises the problem in sports analysis very well in Chapter 4: Light & Dark – so read it if you haven’t yet – there’s quite a lot of rhetoric but the central points are great. This is a quote directly from the book, p132: “defensive actions that can be measured – tackles, clearances, duels – have the feel of one-offs, preventative actions, rather than things that can produce something positive. Ball events are tracked, but things that happen off the ball are ignored. It is far harder to tune in to excellent marking, cutting of passing channels and wonderful positioning.”
Understanding how to credit and apportion defensive work is a troublesome issue in team sport analysis, and many professionals are actively working to understand and model this in order to improve our knowledge of what makes a successful defender. How do you count actions that do not occur?
Today I visited the team at Kickdex, and I was suitably impressed by their approach to football analytics – indeed what I’m now calling ‘the moonwalking bear problem’ was something we discussed and it is something they are actively working to solve. I’m a sucker for the application of any form of science in sport but in my humble opinion Kickdex are looking at technical challenges posed by football in a way that few other people I have met in the industry are. Slightly off topic now but incidentally network theory analysis is another area that we discussed – and I have since been helpfully informed that it is a fairly well-tracked field in the application of science to sport – but nevertheless I feel it is important to highlight a couple of excellent studies on the subject (one of which was co-written by a founder at Kickdex):
Thanks to Paul Power (@counterattack9) for the second link.
If Opta stats don’t give us direct information about the moonwalking bears in football then how can we measure their effect? There is certainly some thinking to be done around this and probably a lot of inferences need to be made from the data we do have. I liken this to a kind of dark matter problem that you might need a theoretical astrophysicist to investigate! Incidentally, the co-author of the first paper linked above is a theoretical physicist called Hugo Touchette. And the director of research hired at Liverpool FC last summer (Ian Graham) has a doctorate in the subject. Clearly, even these fields are attracting the interest of football analytics and vice versa.
I’d personally love access to off-the-ball player position data (and a long, long time to look over it) but that’s something that I’m reliably informed only Prozone and the clubs who install the camera systems have access to in the EPL. Clubs almost always go down the road of keeping information proprietary, which is understandable, but it’s unclear if they have the right expertise to benefit from the information (perhaps Liverpool / Ian Graham excepted). So we need to find a workable solution to understand defence – in much the same way that our current array of stats have shown to be effective at shedding light on offensive performance.
On a separate occasion I have also had confirmation that clubs do indeed tend to have more difficulty in obtaining reliable statistics for central defenders and defensive midfielders, an area that I have also experienced odd results from my player analysis modelling – e.g. in the 2011/12 season, in some of my earliest analyses, I curiously rated Clint Hill as one of the best defenders in the league (fortunately my model has changed a fair amount since then).
So the race to find good data for defenders continues and the significance of inaction on the field continues to confound us. The analytics community has taken great leaps forward for attackers but we are still some way short of tailoring this analysis to defenders. Meaningful information in this area is in high demand!