Like everyone else on The Griffin Park Grapevine fans’ message board, I avidly looked forward to the next contribution from ade211. He wrote exclusively about statistics, analytics and even mathematical modelling, subjects of massive relevance to all Brentford fans nowadays . For someone that struggled with the intricacies of Maths O Level far too many years ago this could have been a step too far for me but ade211 always made a difficult subject fresh and interesting. Rather than just concentrating on and getting bogged down by the actual numbers, he instead demonstrated how to interpret them, what they really meant and the benefit they could bring in terms of understanding the game and how players perform.
What’s more, it was obvious that he was a man of great intelligence who had an open mind and was keen to embrace new concepts and ideas.
His posting were clear, concise, illuminating and often self-deprecating and witty. He never patronised his audience but he was a wonderful teacher with his obvious bubbling enthusiasm for the subject and clarity of expression. In short he was an absolute jewel.
A couple of weeks ago I approached him and asked if he would be prepared to write a longer article on the subject which would serve as a general introduction to the convoluted word of statistics, analysis and mathematical modelling and how best to gain a competitive advantage through their usage. He immediately agreed but stated that he had not been in the best of health and it might take him a little time.
I was therefore terribly shocked and saddened to hear about his sudden death yesterday and only then learned that his real name was Professor Adrian Woods. I contacted his son Anthony to discuss the article and he kindly gave his permission to use it so I am enclosing it today as a heartfelt tribute to a man who I never met but greatly admired.
It is only recently that the application of statistics to football has become widely known and broadly accepted. Today clubs purchase data from companies like Opta using back room staff to analyse the data. If you look at the back room staff involved at say Chelsea or Manchester City you will see the massive resources they have in this growing area. Brentford is now investing heavily in both the acquisition of data and the staff qualified to analyse it and has quickly gained a reputation as being in the vanguard of this move towards the use of stats to help predict performance.
While most football supporters know about the use of statistics, many remain confused about how they are best applied and are sceptical about their value. One of the key objectives is to reduce a manager’s and coaches’ subjective evaluation of a player and replace it with a cold, clinical and more objective one based on statistics. So how does this work? A simple illustration may help to demonstrate.
For players it is possible to report on their shot accuracy as well as the number of shots they take. Let’s take two Brentford players as an example. Last season Andre Gray took 2.68 shots every ninety minutes he played with an accuracy level of 54%. Alan Judge in comparison took 2.35 shots every ninety minutes he played with an accuracy of 38%. Is it possible to delve further into this?
Of Gray’s 2.68, 2.37 were within the box. Of Judge’s 2.35 shots only .65 were within the penalty area. So the difference between their accuracy may be explained in part by the fact that Gray takes more of his shots in the box ( 88%) than Judge does (28%).
Now if we look at Stuart Dallas to we see that he had a shot accuracy of 55%, took 1.74 shots per game with half in the box and half outside. In terms of goals, Gray scored .43 goals every ninety minutes he played, with 93% of them scored in the box. Judge scored 0.09 goals every ninety minutes he played with 67% within the box. In fact this is misleading for Judge as he only scored three goals in total of which two were in the box. Dallas scored 0.24 goals per ninety minutes he played with 50% scored from inside the penalty area.
This exercise could be repeated for each player, broken down by home games and away and for different parts of the season. These are objective statistics, However they still need interpreting by experts to uncover any coaching issues that may arise. Other statistics could also be looked at, such as passing, tackles and chance creation, in order to give a fuller picture of a player’s contribution to the team. It is interesting to note that Judge created 2.41 opportunities per game compared to 0.98 for Gray and 1.31 by Dallas. Judge was also fouled more often per match than both Gray and Dallas. The figures are 2.09, 0.71 and 1.23 respectively per ninety minutes
The data analysis used by Matthew Benham and his analysts uses very good data, analysed by excellent statisticians to provide the manager and his coaches with insights into players that are objective and not biased by subjective opinion. The profiles of Judge, Gray and Dallas used data in the public domain and even then used only a small part if it. Brentford have access to much more in-depth data and have the people skilled in analysing it. This should at least in the Championship provide the club with a competitive advantage.
Using statistics guards against us suffering from confirmation bias. This is when we believe something to be true then filter evidence to support this viewpoint. We tend to remember instances when an event supported our view and forget or undervalue evidence when it does not. My own favourite example is from watching pundits on television argue for putting a man on the post at corners. Now I have no idea if this is a good or bad thing to do, but I do know that most corners do not result in goals, only a very few do. Yet by selective editing it is possible to find the minuscule number of corner out of all the corners taken over a weekend where a man on the post would have prevented a goal. These then get shown on ad infinitum on television to substantiate an incorrect assertion. In an entire season what percentage of goals from a corner would have been prevented by a player on the post, and if the player had been deployed elsewhere would this have prevented more goals, or if a player had been sent forward at corners as Brentford invariably do, would the team have scored more goals from opposition corners ? I have no idea of the answers but I think many suffer from confirmation bias on this topic, remembering dramatic goal line clearances and forgetting all the other times it made no difference at all. By looking at the statistics for this a manager or a coach can be better informed on whether to use a player this way or not without being influenced by their own confirmation bias.
If what other ways can statistics can be used?
The most common use of statistics in football is to assist in the analysis of a player and his value. If the market for players was efficient every player would have a value that accurately represented their actual worth to their team. But as we know this is not the case. What statistics on players help to do is to identify players that the market otherwise ignores or undervalues. In the past scouts would spend time going to games, watching players to see who might represent good value. By using statistics a wider field of players can be searched at a much lower cost and the subjective element in player acquisition minimised. Also scouts watching a player in the flesh may by influenced by chance. They could, by chance, pick the rare occasion when he has a great game. The opposite could also happen. Now players can be identified from their statistics, drawn from a lot of games, before being watched. Looking at the players we have acquired recently clearly demonstrates the work that the background analysts have done in identifying players.
From what Matthew Benham has said, the model used by Brentford has propriety algorithms embedded in it to make the choice of player even more efficient than by our just using other more readily available statistics. As this is not in the public domain given its commercial sensitivity we do not know what it consists of. At the centre though is the certainly that subjective assessment on potential players has been substituted by more objective analytical tools. The players we have acquired recently are the fruits of this approach. We now search a much wider area than others in our league and it is safe to assume that we consider many more players in order to find latent value. In the future other areas of Europe will be searched I suspect, as well as the USA and even Asia. Benham is apparently keen to establish links with the KGH Sports Football Academy based in The Gambia. So perhaps The Gambia will be the next place Brentford looks for players!
Another way statistics can help is by giving players and coaches data on areas of weakness that need attention, as well as ensuring that strengths are maximised. This allows more individual sessions to be planned and their results monitored in game situations. The use of a player’s heat map can show whether they have had the discipline to play exactly how and where the manager wanted. Deviations from this can be discussed and steps taken to rectify errors in positional play. A player’s statistics can also be used to assess whether a player’s impact is declining in so a plan can be drawn up to replace him and also help him find a new, more suitable club to join.
A lot of us, I suspect, will find the use of stats in player acquisition and player improvement not that controversial or difficult to assimilate. Other aspects of the use of statistics most certainly are.
There is a wealth of statistical analysis based on large amounts of data that look at, for example,the probability of scoring from a corner, or from a cross or from shots outside of the box. Most find the low probability of scoring from these quite surprising. With the appointment of a specialised set piece coach our success rate should improve to become, hopefully, above average for the league. One early and well known piece of statistical work was undertaken for Manchester City regarding corners. The manager felt they were not scoring enough from corners. The statistics drawn from across the Premiership found that in-swinging corners aimed at just past the penalty spot had a higher probability of leading to a goal. Even though the manager favoured away-swingers, in-swinging corners were used and more goals occurred. Work also has been done on the kilometres covered by a team in a match. The consensus being, at present, is that the team covering the most ground is more likely to win.
One observation that causes controversy can be expressed by ‘ goals wins matches but defences win leagues’. This emphasises the importance of keeping clean sheets. Under Warburton Brentford played an expansive style but some have argued that defensive weakness caused by over-attacking left us short at the back, resulting in losing points and detracting from our chances of promotion. In about 23% of our games we kept clean sheets, this compares badly to the near 45% of matches in which Middlesbrough kept a clean sheet. Two of the promoted teams, Norwich and Watford, managed 31% and 33% respectively.
One area that has attracted serious work is penalties. Several studies have shown that the keeper staying still in the middle of the goal increases the chance of the penalty being saved. The reason why keepers dive is that we all prefer action to inactivity. So there is a bias at work which psychologically nudges the keeper to dive. Interestingly the taker should focus solely on where he wants the ball to go. If he looks at the keeper in the moment before the kick he tends not to hit the ball cleanly and often it goes high or wide of the goal. This gaze effect has been documented in other sports. Just having someone stand in the middle of the goal, even if not a professional keeper is often enough to disrupt the taker’s ability to strike the ball correctly. I have to say I find that hard to believe but there are several pieces of work on this now that substantiate this view.
At present the Holy Grail is to develop statistics that predict accurately how many goals a team will score in a match. People working on this for obvious reasons are loath to put this into the public domain. Also predicting match outcomes has produced a lot of work of which only a proportion reaches a wider audience There are also models where league position is predicted. Typically these look at wage bill, then injury, then suspension, then managers. The remainder being luck. Running these models from what can be ascertained, means that after about 20 games luck has pretty much evened itself out and a team’s league position reflects the other factors. Some have put the manager’s impact on performance at only about 20%. Evidently Brentford has a very good model for this which was used to identify candidates for the Head Coach position. It may even be able to factor out random events in order to uncover how well the team is doing and if random events had a neutral impact on performance. This mitigates against taking rash decisions based too much on luck and random events.
The experiment being undertaken by Brentford is actually not so radical. The owner is a leading figure in football odds-making, building up resources over several years to produce models that predict match outcome better than this competitors. What I think is new is that he actually understands the statistics and what they can and what they can’t do, as well as having some of the best models available and, crucially, some of the best people to analyse the statistics. He is an astute businessman who has made a considerable personal fortune out of this. In doing so he has acquired excellent skills in running a business. One skill he clearly demonstrates is his strategic ability. Since it become public knowledge exactly how he wanted the club to be run, new managerial and coaching staff have been employed as have new players. These have not been acquired overnight but are the clear end result of having a vision and from that a plan. His majority shareholding in FC Midtylland and the way he has tested his approach there with great success also demonstrates his single-mindedness and determination to break new ground. He has succeeded where others failed in the imminent building of a new ground, something the club has been trying to achieve for years. Again part of his long term vision.
On the whole we are now seeing the outcome of decisions made quite a long time ago. At present he, his board and managerial team are planning for what will be required in the Premier League. Just as players emerged from outside of our radar I suspect this will happen again. The use of better statistical analysis will be introduced as will other more specialised coaches and analysts. At present we supporters are focused solely on the season ahead. I reckon he is thinking five years ahead. Just as with hindsight we can now see what he has been up to for the last few seasons, in five year’s time individual decisions will form a coherent pattern to solidify our place in the Premier League. We may struggle at times to see the big picture but there will be one.
While he is the owner of Brentford I can’t see us going back to the ‘good old days of jumpers for goalposts’ and day trips to Hartlepool where the ball is just belted up and down the pitch.
RIP Professor Woods and our sincerest condolences to his son, Anthony, and all of his family,