By Simon Kuper
When we first published Soccernomics in 2009, we didn’t even have a chapter on match data (stats such as duels won, interceptions, chances created, etc.). Most clubs then didn’t yet take match data seriously, and nor did we. When we published a new edition of the book in spring 2012, we had a long chapter on match data. That’s a measure of how much things had progressed in just three years. Stats are changing football in all sorts of ways. But talking to data analysts at English clubs, I get the sense that one big thing is still missing: stats aren’t yet helping clubs to identify which players to sign.
In the game on the field, data analysis has made the most difference when it comes to set-pieces. That makes sense. A corner, penalty or free-kick, is when football stops for a moment and becomes a static tableau that is fairly easy to analyse – rather like a pitch in baseball. I’ve written elsewhere about how Manchester City’s data analysis of corners helped them win the Premier League (see http://www.ft.com/intl/cms/s/2/04e0e834-e6c5-11e1-af33-00144feab49a.html), about how Chelsea’s analysis of Bayern Munich’s penalty-takers helped decide the Champions League, and about how we at Soccernomics very nearly helped decide the last world cup final (for that you’re going to have to buy the 2012 edition of the book. Get it at http://www.amazon.com/Soccernomics-Australia-Turkey-Iraq-Are-Destined/dp/1568587015/ref=zg_bs_16638_6 ). The next step must be that clubs ban players from blamming direct free-kicks hopelessly into the crowd, and force them to pass the free-kick instead. Statistically, passing would make much more sense.
But when people in football first read Michael Lewis’s book Moneyball, they were hoping for something more. Like Billy Beane of the Oakland A’s in baseball, they wanted to use data to find players who had been undervalued. Beane’s statisticians had used novel stats like on-base percentage to find good baseball players wrongly overlooked by other major-league clubs. Were there key stats like that in football?
Well, if there are, we haven’t found them yet. John Coulson, head of professional football services at the data providers Opta, told me in 2011 that data played a big role in recruitment at only “four or five teams” in the Premier League.
Stats occasionally inform a club’s transfers, but rarely drive them. A rare case of stats in the transfer market was Bolton’s purchase of the 34-year-old central midfielder Gary Speed in 2004. On paper, Speed looked too old. But Bolton compared Speed’s physical data to that of top-class midfielders then at their peak, like Steven Gerrard and Frank Lampard, and came away impressed. Speed matched up to his younger peers, and his data hadn’t deteriorated with age. The data turned out to be a good guide: Speed played for Bolton till he was 38. (Horribly, he committed suicide at 42.)
But on the whole, the quest to assess players using data has failed. When I visited Manchester City’s training ground this summer to meet some of their army of number-crunchers, the club’s strategic performance manager, Simon Wilson, told me: “We’re looking at trying to find the players who are undervalued. A bunch of people have tried by now. It’s not as easy as we thought when Moneyball came out.”
Now City are trying to solve the puzzle by throwing open the match data to all the hobbyists out there (often top statisticians in their day-jobs) who dream of doing a Moneyball of football. (You can find the match data and City’s explanation of their project here: www.mcfc.co.uk/mcfcanalytics).
By the end of 2013, I suspect there’ll be progress to report.
I suppose it makes sense that soccer needs to find its Bill James before it can find its Billy Beane.
Nice article and agree that – when it comes to match analysis – it’s harder to identify KPIs owing to the validity of the game, and easier to make a difference at set-pieces (in swinging corners much more effective than out-swinging as an example).
We’ve also identified individual player KPIs that could be used for player recruitment (and/or managing your existing assets)… a basic example when trying to find under-valued strikers: goals are an obvious KPI, yet goals-scored can be misleading (players in the better teams will score more goals, owing to the support system in place) so you can use a more intelligent metric such as ‘goal-expectancy’ which looks at the number of goals a player scored Vs the number they were expected to score (based on the type of chance and where on the field). Those with a high goal-residual (scored minus expected) may be over-performing and undervalued.
So I think that we do have KPIs for individual players (above is merely one example) and therefore player recruitment, but the bigger challenge for me is how the player will adapt to a new team – i.e. just because a player performs well at x club, obviously doesn’t mean he will at y club). Therefore, the next challenge for technical player recruitment is to improve the understanding of the teams’ contribution to the individual player’s performance and find a metric for adaptation (which could be the ability to adapt both on the field – player style etc – but also off it – culture, language, lifestyle, family etc)
Twitter: @BlakeyGW