A lot on my mind on analytics this week, perhaps with all the end of year crap in the football universe to digest. So I’m breaking this up into chunks.
Please visit the Soccer Analytics Forum
James Grayson set this up after I wrote a post musing on the idea of having a place where analytics studies and subjects could be discussed openly. I initially thought it could be exclusive and snobby to stop random people from posting up crap, but I think James’ emphasis on making it as open as possible was the smart way to go, as it at least gives it a broader reach. If you’re interested in this, please post up articles, questions, links to databases, whatever you like. My own contribution has been decidedly lax lately, so I will change that as of today.
Anyway, I’d really like to see it flourish in the New Year, and that requires some added voices. I’ll be doing some behind the scenes work to spread the word, and will try to use it as a basis for future analtyics columns. OKAY?
Prozone’s “Player Archaeology” Concept
Prozone’s Business development director Blake Wooster wrote a small summary of an interesting project over at Prozone this week. Last week, I mused openly about the exciting and difficult challenge of parsing out individual player metrics, paving the way to perhaps one day allowing laymen and experts alike to assess player potential with one or a set of metrics.
Wooster too is excited by this “undeveloped” approach, and writes that, “It seems obvious to look at past performance in an attempt to identify the factors that lead to success and indicate likely future development, but football’s short-termism rarely affords those involved in day-to-day analysis the time to indulge in retrospective longitudinal research.”
To address this issue, Prozone came up with a means to trace a player’s development throughout their career. It’s ingenious as it is problematic (which we’ll see in a moment):
Using the term ‘Player Archaeology’, we’ve used the example of James Milner, currently of Manchester City, to illustrate how it’s possible to use past performance to unearth key moments in a player’s career trajectory. Since making his Leeds United debut at the age of 16, Milner has played for three other Premier League teams and switched clubs on five occasions. With a long and transient career in the top flight, Milner provides us with a rich narrative to assess in terms of performance development.
Predominantly playing as a wide player, we can assume that coaches generally look for Milner to exhibit playing characteristics such as showing for the ball, crossing, passing and shooting. In the interests of a concise analytical approach, our analysts have synthesised these metrics into one stat: Offensive Efficiency (the number of successful passes, successful crosses and shots for every pass received). As a result of having 10 years of technical data, we’ve been able to plot Milner’s Premier League performance across the last decade.
The method itself very smartly addresses the issue of how to separate out an individual player’s contribution to the team, with a per-possession/successful action measurement. There are some significant issues with this approach, however. For example, successful-pass per possession rates could be impacted by a poor team, as players could be less able to get into good positions to receive them. The same goes for shots and crosses (teammates not able to draw defenders out of position for shooting opportunities, or get in positions to receive crosses).
Even so, the results, which can be seen on the site, are very interesting. First, there is a significant upward career arc for Milner, and second, there appears (as the study’s author notes) to be an adjustment period each time he switches clubs. Whether this adjustment period can be found in other elite players is something that should be explored further.
Wooster acknowledges that this method lacks predictive power, and certainly the gulf in improvement from one club to the next in Milner’s case might undercut the notion that these performance metrics are somehow absolute, and innate. Still, it’s a good start in looking at a very difficult problem in football analytics.
Jonathan Wilson’s year-end tactical review under attack
His latest year-end summary of tactical trends on the Guardian have been met with some criticism on Twitter, for several different reasons. Some have accused him of rehashing trends that were largely on the rise at the end of 2011.
Others however have accused him of simplistic generalizations, as with his assertions that the importance of the striker is on the wane, which would conflict with the importance of Robin van Persie at Manchester United or Ibrahimovic at PSG or Falcao at Atletico and etcetera.
I think my problem with the piece is one of form. Football tactics are so diffuse and context-specific that to point to any one overarching trend or movement is to risk gross oversimplification or omission. The best kind of tactical writing in my opinion is that which is anchored in specific games or moments; measuring their effect however on future trends presents a major challenge.
That’s why I think too that for too long, tactical analyses have given way to far too much subjectivity. If one, amazing team like Barcelona favours a particular approach, it becomes very easy to read it into the approach of teams with ostensibly similar tactics. I also think that many tactical assertions are made without recourse to empirical evidence.
To some degree, data-driven analyses can help counter that trend. For example, Wilson states that “highest level the majority of games are won by the team that best controls the ball.” He’s referring to possession, but possession itself is far more fuzzy and less definitive than a metric like total shots ratio, which indicates that a side is effectively using their possession to take more shots and concede fewer, a metric with solid predictive power as far as final table position.
What’s interesting is that the data, particularly on Barcelona, confirms some of Wilson’s impressions whilst contradicting others. He writes for example that “Tito Vilanova’s team looks more direct than Guardiola’s and is perhaps a touch more functional,” an assertion that is borne out a bit by the data. The team this year, as compared to Guardiola’s, has a lower total shots ratio (.585) in the league, meaning they’re taking fewer shots and conceding more than last season, when the club averaged above .650.
However, Barcelona’s La Liga PDO—a metric which adds their shot percentage to their save percentage—is 1135, a reflection both of luck and the incredible efficiency and skill of their best attacking players, with Messi leading the way. Whether a false 9 or not, there is a lot of evidence that the club’s reliance on their star forward, who’s scored 25 of the club’s 54 league goals.
In any case, sometimes the metrics can bear out some nuance that broad impressions cannot.