Evaluating Possession Losses using El Clasico
Providing context and evaluating possession losses in a football match, using el Clasico as an example.
Turnovers are a tricky subject in football. Whereas in other sports, turnovers can accurately be deemed as costly, it’s not that simple in football, where spells of possessions are longer and games are relatively slower. Although we constantly see player-specific stats such as pass completion rates and dispossessions, such indicators are stripped of context.
Some players are more inclined to take riskier decisions by attempting to create something, thus lowering their pass completion rates and increasing their turnovers. While in most cases losing possession is far from ideal, there are certain situations where a player, through seemingly wasteful action, can unsettle the opposing defense and create openings despite directly losing the ball.
In this article, we will attempt to put unsuccessful passing in greater context, using the recent el Clasico as the case example. Throughout this article, we will be referencing the possession value model developed by Laurie Shaw’s (@EightyFivePoint). In the simplest terms outlined by Shaw, possession value is the probability that a possession will result in a goal.
Successful Passes:
Firstly, we will look at successful passes and see which players generated the most value from their passes. Progressive passes (purple) represent passes that increase the possession value, while regressive passes (yellow) represent passes that decrease or keep the possession value equal. Figure 1 shows the possession value leaders for Real Madrid.
Unsurprisingly, Toni Kroos leads the way for Real Madrid. He is a wonderful passer and crucial to moving Real Madrid up the pitch and allowing them to progress the ball. His score of 0.42 is also a match high. Through his passes, Toni Kroos increased the chances of Real Madrid scoring more than any other player on the pitch.
Next up, FC Barcelona. Figure 2 shows the possession value leaders of FC Barcelona.
Lionel Messi, unsurprisingly, finds himself close to the top, while Jordi Alba was the highest-ranked player for FC Barcelona. Most of Barcelona’s play flows through Messi, and the Catalans made a point of targeting the left hand side, where Alba is their main outlet.
Unsuccessful Passes:
Now, what about misplaced passes? Surely, not all are equal. Some provide more value than others, while not necessarily rebounding negatively on the team that commits them. To calculate this, we looked at all unsuccessful passes and calculated their possession value added based on the start location and intended end location (corners were excluded as they might skew some of the data). Like successful passes, misplaced passes are divided into progressive and regressive. Let’s start by looking at Real Madrid. Figure 3 shows Real Madrid’s possession value from their unsuccessful passes.
Next up, we looked at the average possession value lost for their own team, as well as the possession value given to the opponent. Average possession value lost was calculated using the possession value of the pass start location and averaged for all passes. For example, if a player misplaces a pass that had a PV of 0.01 at the start location, he or she would be credited with PV lost of 0.01. The higher the value, the more value lost for his own team. For opponent possession value, we looked at the opponent PV at the intended end location of that pass. For example, if a player passes the ball to an area where the opponents PV is 0.01, then that player is credited with an opponent PV of 0.01. The lower the value, the less costly the turnover is. Figure 4 displays these values for Real Madrid.
We now have enough information and context to evaluate a player’s possession losses – how valuable a given loss and how costly. Vinicius Junior had the lowest pass retention for Real Madrid, misplacing the most passes. However, when we look at the above graph and table, we see that he did so by attempting to increase the probability of his team scoring, while his misplaced passes did not yield much value to Barcelona. (This could also raise questions about decision making in certain cases)
The same was done for FC Barcelona. Figure 5 shows the same table for FC Barcelona and Figure 6 shows the possession value leaders from unsuccessful passes.
Let’s take the example of Lionel Messi, who had a pass completion rate of 89.2% and lost the ball more times than any Barcelona player (excluding corners) with the exception of Jordi Alba. Despite losing the ball relatively frequently, Messi attempted to increase his team’s probability of scoring more than any player on the pitch. As seen in the table, the PV lost by Messi did not generate much value for Madrid . Messi’s pass map in Figure 7 makes clear that most of his passes looked to progress the ball to dangerous and high value areas.
To conclude, we have attempted to contextualize possession losses through passes using a possession value model. The idea was to see why players are losing possession and whether such losses are costly or valuable to their teams, giving new meaning to high risk-high reward players. Some limitations/areas of improvement with this approach:
Only passes were included. Other actions such as carries are as important and must be evaluated accordingly for a better picture.
The net possession value for unsuccessful passes was based on the intended end location of the pass. Since the pass was unsuccessful and did not reach it’s intended location, there is some margin of error here in the end location.
Similar to the above point, the opponent possession value gained was also based on the intended end location so similar issues as the point above would arise. Besides that, passes may have been intercepted and turned over to the opposition at a different location, so calculating the possession value of the location of the interception might be more valuable than isolating the intended end location. Alternatively, not all unsuccessful passes are turned over to the opposition so it might be better to only include those passes that were turned over through a throw in, tackle, interception and where possession remained with the opposition after that action.
The situation of the possession is not factored. Possession losses and value in counter attacks and possession based attacks should probably be treated differently.
The next action is not accounted for. A lot of value from misplaced passes comes from unsettling defenses and creating good openings elsewhere. At the moment, this framework does not account for that next action. In other cases, an unsuccessful pass might get intercepted but remain in possession with the same team, so an alternative method would be to calculate the PV of the start location of that action rather than the intended location of the pass.
Another way of evaluating how costly a possession loss is would be to look at the next opposition actions and whether anything meaningful occurred.
As is the case with most metrics, comparing players with different roles and positions does not make much sense. Defenders will most likely lose possession in more dangerous areas for their team, while attackers proximity to goal would no doubt play a role in their higher PV figures.
There is plenty of room for improvement. This is the first draft of an idea for contextualizing possession losses and hopefully provides some insight into an alternative method for looking at possession losses.
This idea was partially inspired by StatsPerforms recent changes to their PV model. All PV values were calculated using Laurie Shaw’s possession value model.