Parker’s Possession Problem

One characteristic of ‘Parkerball’ as played by Fulham this season, was domination of possession.

Fulham averaged 61% possession this season (using sofascore based possession stats). They had the majority of the ball in 42 of the 46 regular season games, indeed the only Championship team’s to achieve similar numbers in thee last 3 years were Bielsa’s Championship winning Leeds team (64%) and Slavisa Jokanovic’s promotion winning Fulham team (61%) in the 17/18 season.

In 14 games, Fulham managed what might be classed as ‘extreme possession dominance’, achieving over 65% possession control. That sounds good, but then consider that Fulham only won 3 of those 14 games, and one of those 3 wins was against a Reading side that played most of the match with 10 men (and indeed when Reading visited the cottage and held on to a full team for 90 minutes, Fulham lost the game despite 71% possession).

We can explore this a bit further in the chart below, it’s a little complicated and takes some explanation:

The bar chart shows, on the horizontal axis, Fulham’s possession in a given game.

Let’s focus on the yellow bars first, these show average net xG per game. That is a bit of a mouthful, but it is actually not that complex, net xG is a measure of the expected goals generated by Fulham minus the expected goals generated by their opponent. In other words, the yellow bars represent the average expected margin of victory (or defeat if negative) for Fulham in each game as determined by the xG model.

It’s still a bit confusing, so let’s walk through it: the first yellow bar tells us that: in games where Fulham achieve 50% or lower possession (there are only 4 such league games), xG models expect Fulham to lose by an average of 0.6 goals per match.

In games where Fulham achieve 51% to 55% possession (there are 6 such games in regular season), xG models expect Fulham to lose by a slightly smaller margin, an average of 0.35 goals per match.

In games where Fulham achieve 56% to 60% possession (there are 15 such games in regular season), xG models for Fulham are (amazingly) completely even, Fulham are expected to concede exactly as many as they score!

In games where Fulham get 61% to 65% possession (there are 6 of these), they seem to hit something of a sweet spot, with xG models predicting an average victory margin of 1.2, showing Fulham are very dominant in chance generation in these games.

So far, we have seen that, as Fulham’s possession rises, the results in xG terms improve. Which is not particularly surprising as high possession indicates a level of control in the game.

But the interesting thing is what happens next, as Fulham’s possession rises, the net xG starts to fall again.

When possession is between 66% and 70%, net xG falls to 0.3. And for possession over 70%, it is just 0.1. So, the stats are saying that Fulham are being less successful in games as possession rises above 65%.

Do these xG numbers match with reality? Yes, they do seem to…

The blue bars in that chart above give the actual points earned per game in each of the possession bands. As with the xG, they are generally rising as possession increases, peaking in the 61% to 65% possession band. But then they too fall away. In fact, they fall away remarkable sharply, Fulham still have positive xG in these high possession games, but the points totals are lower, even than when Fulham have sub 50% possession.

This goes back to agreeing with our original observation, showing that Fulham struggle to win games when possession goes over 65%.

What Happens to Fulham in these Games?

I have pulled up the Infogol shot maps for two of the more egregious examples of these ineffective high possession games.

Against Hull at home Fulham had 75%possession and lost 3-0, and against Stoke (away) they had 78% possession and lost 2-0. The circles on the maps show the position of shots, and the size of the circle shows the xG of the chance in question.

In the game against Hull, Fulham managed 8 goal attempts, but you have to look very hard to see them on the map because they were very low xG chances indeed and therefore the circles are very very small! There were a couple of at least visible chances in the Stoke game, but mostly it was miniscule probability long range pot-shots at goal.

Fulham’s opponents meanwhile mount a smallish number of much higher probability chances, typically earned on the counter-attack. In these 14 possession dominant (>65% possession) games, Fulham outshot their opponents by 188 to 122, but their xG per shot was a very low 9.3% of a goal, while opponents were generating 13.8%of a goal per shot in xG terms.  In other words, the opposition shots were, on average around 50% more likely to result in a goal than Fulham’s shots.

The opponents were limiting Fulham to low quality chances, and waiting to hit Fulham on counter-attacks to generate a high value chances.

Is it a problem of tactics?

There is a question as to whether Fulham can do anything about this, do these high possession games happen because of the way the other team set up, or is it a consequence of Fulham being too slow and ponderous in attack, or a combination of both?

That’s hard to answer confidently, but I do note the chart below which indicates a relationship between the Fulham goalkeeper’s choice of long vs short passing and the team’s overall possession stats.

On the bottom (x axis) we have the proportion of goalkeeper passes classified by Infogol as a ‘long ball’. This ranges right from 100% (zero short passes) to 0% (entirely short passes).

We have all seen that sometimes Fulham insist on playing out short from the keeper, even when to do so seems high risk. This suggests a deliberate decision to dominate the ball and possession and this is supported by the chart which shows, in general, the more the keeper plays short, the higher Fulham’s average possession tends to be (shown on the vertical y axis).

The yellow trendline highlights this relationship between high frequency of long balls from the keeper and lower possession.

The chart also shows the outcome of each of the games (colour of the dot) and highlights Fulham’s problematic 65% possession line, above which lie the 14 extreme possession games that Fulham struggled in.

It is also notable that these extreme possession games, and indeed the games with very infrequent goalkeeper short-passes, were earlier in the season, suggesting that Fulham have moved, as the season progressed, to try to be a little more direct with a beneficial impact on points earned per game as a consequence.

Is it just Fulham who struggle with high possession?

I considered ways to test if this possession problem (results deteriorating when possession goes very high) is a general theme in football or a unique Fulham problem.

The first test of this was to compare to Slavisa’s 17/18 promotion winning side which also liked to dominate possession. Like Parker’s team, Slav’s also averaged 61% possession.

They had a similar number of >65% possession games (15), winning 6 of them. Better than Parker in points per game terms, but still hardly promotion form.

I reproduced the xG and pts per game bar chart for Slavs team, and this is shown below.

The pattern is not that dissimilar to Parker’s team, other than an anomaly in the 56%-60% range, net xG and points are generally improving with possession, until the 65% line is crossed at which point both net xG and points per game worsen.

So two Fulham teams seem to have experienced this issue, I wondered if this was a general footballing phenomenon, i.e. teams with very high possession struggling to win.

To look into that question, I got hold of Premier League match data from fbref.com.

From this I looked at results, net xG and possession from all the games from the last 3 seasons (about 1,000 matches).

The blue line shows that, across the circa 1,000 premier league games, there is a simple linear relationship where, as possession increases, so does a team’s net xG (expected margin of victory). There is little sign of that deteriorating as the very high possession levels are reached, contrary to the recent Fulham seasons.

It seems to be simply the case that the more of the ball a premier league team has, the better they do.

OK, but what about the Championship

I undertook a similar exercise looking at 3 years of championship data. The source (Fbref.com) did not include xG in its championship match summaries, so I had to look at points per game against possession instead, and I noticed something fairly extraordinary…

Just like the premier league there is a clear correlation between possession and points earned per game, but in the championship that correlation is negative!!!!

The data is clear, every possession bucket generates a lower points per game than the one below it (except 66-70% but that is off quite a small sample). There is a steady trend here, possession goes up, results go down!

An interesting feature of these numbers is that they contain results pairs: So, for example there are 74 games where a team achieved >70% possession. And therefore, by symmetry, there are also 74 games where teams got less than 30% possession (i.e. the opponents in those same games).

Of the 74 games, those teams with less than 30% possession won 33, while the possession dominant teams won 19 (with 22 draws).

In every set of result pairs, the outcome is the same, the lower possession teams always win more games.

The chart below compares to the Premier League, where the positive relationship between possession and results is clear.

What is going on here?!?!

I find this to be fairly mind blowing as an observation, unlike in the premier league, teams in the championship are more likely to lose the more possession they have.

This is particularly counter-intuitive because a number of successful championship teams have been built around high possession football. This includes both promoted Fulham teams and also West Brom, Leeds, Brentford and last season’s champions Norwich.

I looked at teams from the last 3 seasons who had finished the season with high average possession (and all these teams were quite successful, with most getting promoted, other than Swansea and of course, Brentford). The table shows the results of these teams in the games where they managed more than 65% possession and then ranks them by their success in those games.

The numbers show that Parker’s struggles against highly defensive setups are far from unique. Indeed, only this seasons Leeds under Bielsa have really maintained their league form in these high possession games and as a group these otherwise successful teams lost more than they won in these games.

It also seems to be worth noting that when Fulham beat Leeds this season, they did so by conceding possession as Fulham dropped to 35% possession in that game. Conversely in the reverse fixture at Elland Road, Fulham had 63% of the ball in the 3-0 defeat!

Let’s wrap this up…

At the risk of outstaying my welcome here: let us sum up what we have learned:

  • Parker’s Fulham struggle to break-down very defensive teams when using high possession-based tactics.
  • However, they are not unique in this problem, most possession-based teams have struggled with this in the championship in recent seasons, and only Bielsa’s 19/20 Leeds side has shown real ability to consistently break down such defensive teams whilst also dominating the ball.
  • More direct approaches to attack (including the keeper playing the ball longer more often) may have been a factor in improving results as the season developed for Fulham.
  • Premier league teams appear to have no such difficulty winning if their opponent is willing to concede possession with probability of victory increasing linearly with possession.

I suspect what we are seeing in these numbers is a feature of economics. My hypothesis is that attacking attributes are highly priced in the market for players whereas defensive competence can be bought relatively cheaply.

Consequently, in the financially constrained championship, the most cost efficient team set up is to load up on defensively sound players and then purchase a small number of more expensive flair/attacking players who can be deployed effectively in counter attack situations.

A good thought experiment for such a team is Hull, who, via Jarrod Bowen and Kamil Grosicki delivered a counter-attacking masterclass in the aforementioned 3-0 mauling at the Cottage as part of a compact and organised defensive setup. But once those two players were sold in January, that team quickly fell apart, eventually getting relegated in last place.

In the premier league, resources are not so constrained and teams are free to load up on defense breaking star attacking players. In that league, teams seek possession and the defensive low block is a less viable strategy. There are still some premier league teams, however, Crystal Palace spring to mind, who do still seek a championship style set up built around organised defense and the use of a small number of attacking flair players.

It would be interesting to explore other leagues in the UK and around the world to see which of them also demonstrate the Championship’s negative correlation between possession and wins. My prediction would be that we should see this in quite financially constrained leagues but not in the elite divisions of world football.

For Fulham, and other Championship teams seeking to dominate the ball, I think the lesson is that, when faced by a team intent on hardly touching the ball, simply trying to pass through them will not work (unless you do whatever Bielsa did in those situations), instead a more direct approach may bring more success, despite the impression that one is turning the nature of the game into one that naturally suites ones opponent. I think there is evidence that Fulham did learn this lesson in the second half of last season, and results benefitted accordingly.

Anyway, thank you for staying to the end, I think this piece may have raised more questions than answered, I may return to it in future, but if you do have any questions on the above, please put them in the comments and I will do my best to respond!

EDIT: Update 20/8/20

Several readers (including BenP in the comments here) have put forward the hypothesis that shape of Fulham’s possession relative to results could be a reflection of the state of the games in question. For example, games where Fulham concede early and the other team tries to park the bus, may be likely to generate high possession figures for Fulham (as they try to break down a stubborn opponent) and are also quite likely to result in defeat, given they go into such situations, by definition, behind in the game.

Conversely, where Fulham go ahead, they are themselves likely to adopt a more defensive stance, and therefore build up slightly lower possession.

To test this, I looked at Fulham’s games this season and looked for any correlation between the number of minutes spent with one team in the lead and the possession figures.

This is shown in the chart below and what we are looking for is evidence that the more minutes Fulham spend in the lead in a match (represented by positive values on the vertical y axis), the lower their possession tends to be (shown on the x axis), whereas minutes spent trailing in the game (negative numbers on the vertical y axis) should be associate with very high possession.

Looking at the chart, it is not obvious that such correlation exists, So while I think the hypothesis makes logical sense, it is not apparent in this data set.

5 thoughts on “Parker’s Possession Problem

  1. Hey,
    Lovely stuff. The thing I was wondering throughout was how is the correlation with time and team of the first goal with net xG, Ors and high possession games. We know Fulham’s PTS success when scoring first Vs not, and can recall painful games like Forest, Stoke and even Leeds away where extra possession when chasing game Leads to conceding again. Anyway, Interesting, thanks!

    Liked by 1 person

    1. Ben – for info I have updated the article (at the end) to show the data you asked about. Its not obvious from these numbers that their is a link between minutes spent in the lead/behind and the possession figures from this, although I agree it is a sensible hypothesis

      Like

      1. That’s awesome, thanks. Yes, it’s for sure not clear from that. To see link with original point, Is it possible to colour the marker to represent the game outcome?
        Cheers

        Like

  2. It is a good question, i think your point is that the state of the game might create situations where a team in the lead is less likely to seek ball possession, so by definition a team that has high possession is one that is already losing.

    I will look into the correlation for Fulham between possession and time of first goal to see if there is a relationship!

    Like

  3. Just to note to Ben – I have added the colour coding suggested, losses are shown as red dots. They look fairly spread against the possession ranges to me, but one interesting feature is that they are near the bottom of the chart, which indicates, Fulham spent a lot of minutes behind in those games, ie Fulham conceded an early goal.

    We all know the number about Fulham winning when they score first and not when they concede first, but from this chart it looks like it is probably the case that Fulham probably only rarely lose at all if they are even level at half time. Indeed I would guess from this that Fulham have only lost one or two games when they have been level after 30 minutes.

    Perhaps a future blog on this point!

    Like

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: