In his debut blog for OptaPro, data scientist @WillTGM analyses central defender passing trends in the Premier Leag…… 17 Jan

VACANCY - we are recruiting a data scientist, who will work across @OptaPro & @OptaSuit. Opportunity.… 16 Jan

Fantastic to have @WillTGM join the team. We'll be publishing Will's first blog for OptaPro this week, which explor…… 14 Jan

Valuing passes – @GarryGelade considers how passing performance can be evaluated using a Pass Reward model. Vision.… 10 Jan

The OptaPro Blog

RSS Feed

Welcome to the OptaPro blog, featuring news and analysis from OptaPro's cutting-edge research team.

BLOG: Take him on

In this article Johannes Harkins analyses the attacking impact of a take-on in the Premier League.


Opta defines a take-on as:

An attempt by a player to beat an opponent in possession of the ball. A successful dribble means the player beats the defender while retaining possession, unsuccessful ones are where the dribbler is tackled, Opta also logs attempted dribbles where the player overruns the ball.


Last season in the second league fixture for each side, Arsenal faced Everton following a tight race between the two sides for fourth place in the previous season. In the first half Everton staked a commanding 2-0 lead (which they’d later relinquish). The second goal in particular made for exhilarating viewing. After outmuscling Per Mertesacker to collect a clearance on the right wing, Romelu Lukaku turned and raced up field with Arsenal’s other centre back, Calum Chambers, speeding towards him to close him down. Faced with a number of options, what Lukaku chose to do - take-on Chambers on the dribble - demonstrated just how valuable this action can be in displacing defensive shape. By the time Lukaku had successfully taken on Chambers, Arsenal’s defensive shape was more or personified in the lone figure of Mathieu Flamini, caught in no man’s land between Lukaku and Steven Naismith.

Romelu Lukaku

Stretched between the two attackers, and with limited help, Flamini could do little to prevent Lukaku from feeding Naismith, who dispatched the goal calmly.

While this example shows the best case scenario for a won take-on, attempting to dribble past an opponent just as often results in frustration as a thwarted attempt sees possession head the other direction. The nature of the take-on is that one way or another a player will be briefly removed from the play: a defender has to chase after a player who’s beaten them, or a player must recover after giving possession away. In evaluating whether or not a take-on is a worthwhile endeavour, we should be thinking about the value of both of these outcomes, as well as the likelihood of each scenario.

The trouble is that totals for take-ons give an incomplete picture. Success rates amongst even the five players who attempted the most take ons since the beginning of last season range from Eden Hazard’s 63.4% to Wilfried Zaha’s 43.0%. In order to more rigorously evaluate value created from take-ons, we need to look at other events surrounding the take-ons to add context.

First, let’s look at where on the pitch take-ons occur. This chart shows the frequency of take-ons attempted since the beginning of last year in the Premier League.

Premier League take-ons (2014-16 seasons to date) 

PL Take -ons

Attacking left to right

As expected, volume is high on the wings in attack. This matches intuition as well as a rough internal calculation of expected value; these areas seem like they might be low in cost for losing the ball, and may have high value for winning a take-on. We can also look at take-on win rates to evaluate their efficiency. I left the total number of take-ons in each area as the label so as to allow for evaluation with the context of volume.

Premier League take-on win % (2014-16 seasons to date)

PL Take-on Win Percentage

At first glance, seeing a lot of take-ons won in the defensive half seems a little odd, but it highlights an important issue which is often lurking in analysis of event-based data in football. The issue is one of selection bias. We can only observe take-ons which actually occur, which means the player has already made a decision about whether or not this is an appealing option. We can’t record the times someone forgoes a take-on because of a well-positioned supporting defender or lack of space. This is a tough problem to solve, but it is important in terms of context for viewing success rates.

With this in mind, one thing which does stick out is the relatively low rate of success in the wide areas which have such high volume. This raises the question, are too many take-ons being attempted in areas which are inefficient? To find an answer, we need more than just success rates. We also need to evaluate what it is worth to win a take-on. For the purposes of this article, I’ll be focusing primarily on the direct attacking value of a take-on only.

My basic method for evaluating the attacking worth of a won take-on was to analyse shots that either directly followed or which were assisted by a pass following the take-on. This method isn’t perfect - it’s a rather narrow definition of how a take-on can lead to an opportunity - but it is clear cut. It’s feasible this method will miss some opportunities which were strongly influenced by a player completing a take-on, but it’s unlikely to falsely identify many chances where the associated take-on didn’t hold much relevance.

Now we’ve got a basis on which to assign attacking value to a successful take-on.

Expected goals models are already a well-established way to evaluate a shot’s expected value, so in this context we can evaluate the value of a successful take-on by looking at the expected goal value of the resulting shot.

By looking at the resulting xG generated per take-on, we get a stronger understanding of the attacking value of a take-on from each area; shown in the visualisation below (the total volume numbers are again included for context).

Premier League take-on value (2014-16 seasons to date)

PL Take-on Value

From this perspective the wide areas which were so frequently the locations of take-ons turn out to be relatively low-value opportunities for directly generating chances. This is probably a conservative estimate of value because of the limited definition I’ve set in this case of how a take-on can add value. However, by extending the methodology of contextualising events by their surrounding counterpart events, a more context-rich sense of value could be added.  

We can also apply this method to individual players to look at their expected attacking value from different areas relative to average rates, and in juxtaposition with their volume. For instance, charting Eden Hazard’s attacking value from take-ons makes him look downright menacing around the penalty area.

Eden Hazard take-on efficiency (2014/15)

Eden Hazard

Compare this to Ross Barkley, he of 128 attempted take-ons last year, who should perhaps elect to find a pass a bit more often.

Ross Barkley take-on efficiency (2014/15)

Ross Barkley

There are some obvious and non-trivial caveats to this analysis. The narrow definition of a dribble’s value assumes to purpose of a take-on to be the generation of a shot. Clearly this isn’t the case for all take-ons, so suggesting, for instance, that players cease to attempt take-ons in their own half on the basis of this analysis would be flawed. Another way to approach this analysis would be to loosen the definition of value and look further down the string of events following take-ons. Furthermore, we haven’t considered the defensive implications of surrendering possession in the course of losing a take-on. Clearly, there’s a lot yet to be unearthed about the risk-reward trade-off of an attempted dribble, but the case study here shows how a relatively straightforward linking of how two or three event types can serve to add a great deal of context to a single counting statistic.


The graphics in this piece were influenced by the style of Kirk Goldsberry of FiveThirtyEight

Posted by Johannes Harkins at 00:00
Tags :


Post a comment