On a number of occasions recently, Phil Parkinson has quite rightly pointed out that his Sunderland side’s expected goals (xG) is excellent. So good in fact, that Sunderland would top the xG table - if there were such a thing.
In his post-match interview with the club website following the defeat to MK Dons, Parkinson again mentioned it when he said:
Our expected goals is a lot higher than the actual goals we’ve scored, and that’s because of the quality of the chances we’ve created.
However, on this particular occasion, the manager was wrong about the xG stats because Sunderland ended the game with 0.71 to Milton Keynes Dons’ 1.47.
It won’t have escaped any fan’s notice that over recent weeks - most clearly in the period following the Crewe home game - defensive performances have declined. What was only a few weeks ago hailed as the ‘best defence in England’, has now conceded twelve goals in seven games in all competitions.
If we look at just the league games, it’s eight goals in five, resulting in goals conceded per game of 1.60. But Sunderland have well documented problems with defensive injuries. That, combined with the influence of international call ups, certainly hasn’t helped recent displays. Perhaps it is unreasonable to be too critical. Although the manager doesn’t do himself any favours when he plays O’Nien at centre-back while Sanderson sits on the bench as an unused sub.
Looking at the season as a whole, the xG per game for and against (which I will refer to from hereon as xGf and xGa) is 1.83 and 0.90, respectively. Those are good numbers. Were the side to convert them into actual goals, they would be a very good bet for automatic promotion. But in reality, Sunderland’s actual goals for return is 1.36 compared to 0.82 against. Not bad numbers, but I’m not putting a tenner on automatic promotion on the basis of them.
If I were to ask you to name the current League One side that presents the most dangerous attacking threat, would your answer be Sunderland? I doubt it. You might say Portsmouth, Peterborough, Doncaster, or maybe Lincoln. But not Sunderland. Attacking threat is not what we do. Yet that is what the xGf number is suggesting.
Why such a disconnect between xGf and our perception of what we are seeing?
Below is a scatter plot of attempted shots on the x-axis and expected goals on the y-axis for every League One side.
In the main, the trend is linear from the bottom left to the top right. That is, low shot count correlates with low xGf and vice versa.
Two sides stand out as outliers, however. Namely, Lincoln City and Sunderland. Both sides have much higher xGf than the number of shots they are attempting suggest they should have. One might expect Lincoln to have an xGf in the range 1.05 to 1.40. And Sunderland should expect 1.25 to 1.55.
However, Lincoln’s xGf is 1.76 and Sunderland’s is 1.83. To get to the bottom of what is happening, we need to go deeper.
Below is a similar scatter plot, but this time shots on target replace shots on the x-axis.
Notice that Lincoln’s datapoint shifted to the right, indicating that they are getting a good portion of their shot attempts on target. Sunderland’s datapoint, on the other hand, shifted left, indicating that they are doing less well in that respect.
The xGf of Lincoln now makes more sense and falls broadly into alignment with the trend for the league. With their SOTf being 5.09, the xGf figure of 1.76 seems about right.
On the other hand, Sunderland’s xGf now appears to be even more of an outlier. From 4.00 SOTf per game, xGf per game of 1.83 is being produced. The regression line suggests xGf should be in the range 1.00 to 1.35. This doesn’t add up.
To make sense of this situation, I took the top four xGf sides in League One (namely Sunderland, Lincoln, Peterborough, and Portsmouth) and examined in detail every shot attempted. From a total of 565 shots, each of the following was recorded:
- Shot on or off target
- Head or foot
- Free kick (direct or within three touches after the kick)
- Corner (direct or within three touches after the kick)
- Distance from the goal when the ball is struck.
- Count of opposition players between the ball and the goal as the ball is struck.
- Whether the shot resulted in a goal or not.
Starting with some overviewing, the stacked bar chart below gives the total shots per 90-minute period for each side subdivided into shots off target (black) and shots on target (green).
Portsmouth have the most shot attempts per 90 (13.29), followed by Peterborough (12.87), then Sunderland (11.43) and lastly Lincoln (10.19). The proportions of shots each side are getting on target are 44%, 39%, 37% and 31% for Lincoln, Peterborough, Portsmouth, and Sunderland, respectively.
31% of shots being on target for Sunderland is poor compared to the other three. Perhaps Phil Parkinson is right when he says the only problem is poor shooting and finishing?
Not quite. Let’s continue digging.
Next I show the proportion of shot types for each side considered.
We start to see some clear differences between the shooting styles now. Lincoln and Peterborough favour creating chances involving shooting with feet, while Portsmouth and Sunderland proportionally pursue tactics resulting in headed shots.
This probably shouldn’t be too surprising since Parky has made no secret of the fact he believes in getting his Sunderland side scoring from crosses.
The next chart shows the proportion of shots by source.
For all four, as you would expect, the most common source for a shot attempt is from open play. But both Lincoln and Sunderland appear to rely on corners more than the other two.
Now I expand a little on shot types and sources and examine the proportions that are on target. Starting with head and foot shots.
Of the four sides considered, Sunderland rank third for getting headed shots on target. Lincoln do very well indeed with 50%, followed closely by Portsmouth with 45%, then Sunderland on 35% and finally, Peterborough on 29%. We saw previously that the two sides who seem to actively pursue headed goals are Portsmouth and Sunderland. Portsmouth appear to be significantly better at getting them on target, however.
Sunderland rank fourth for getting foot shots on target. Lincoln again do well getting 43% on target, followed by Peterborough with 40%, Portsmouth on 35%, and then Sunderland with 31%.
Next up is the proportion of shots that are on target separated by their source.
All four side have got 100% of their penalties on target, as you might expect. Sunderland are the least effective from free kicks, and from open play, but do a little better from corners. Lincoln and Peterborough are the most effective from open play.
What I hope you have picked up from reading this far, is that Sunderland aren’t great at converting their shots into shots on target and, consequently, into goals. Whereas the other sides (Lincoln in particular) are much better at it.
Observe the next chart, which shows box plots of the distances from goal at the moment the ball was struck. For the uninitiated, box plots are a handy way of visualising and comparing the spread, skewness, mean, median and range of a set of numbers. The simplest way to think of it might be to look at the box parts of the box plots and think of that as being the area from which most shots are attempted.
For the first time in our discussion, we see Sunderland come out on top for something. They are shooting around four yards closer to the goal, based on the median, compared to Portsmouth and Peterborough. And about two yards closer than Lincoln. Keep this information in mind as I continue.
Being close to the goal when a shot is attempted is a good thing. But how many opposition players there are in the way matters too. The next chart provides box plots for a metric which has a name that just rolls off the tongue - ‘players between ball and goal’ (PBB&G).
The chart actually shows ‘opposition outfield players between the ball and the goal’ as the shot is struck. What this shows is that Peterborough and Lincoln are shooting with fewer opposing players between the ball and the goal. While Sunderland and Portsmouth are shooting when there are more.
Proportions at each PBB&G count are provided in the next chart.
Both Lincoln and Peterborough create shooting opportunities that leave a clear path between the attacking player and the goalkeeper around 30% of the time and one opposing player a little less frequently (27%). What this means is that they are setting up shooting opportunities from which the attacking player has only the goalkeeper to beat roughly 30% of the time.
Portsmouth create shooting opportunities with a clear path to the goal less often (19%) and Sunderland do this least often of all with 16%. This tells us the Portsmouth, and especially Sunderland, are setting up far fewer chances of the type enjoyed by Lincoln and Peterborough.
The approximate probabilities of getting a shot on target and of scoring a goal at the various PBB&G counts are shown in the table below.
Approximate shot on target and goal probabilities
From these probabilities, it is possible to calculate an estimate of the number of shots on target - as well as the number of goals - each side should have expected to get by this stage of the season. The results are provided in the next two tables, starting with expected and actual shots on target and then expected and actual goals. (Note, expected goals is the number of goals predicted by the probabilities and is not to be confused with xG).
Estimated and actual shots on target
Estimated and actual goals
Pleasingly for our analysis – though clearly not for Sunderland’s chances of automatic promotion – the estimates are good. Sunderland are performing about as well as can be expected given the characteristics of the chances they are creating. In fact, they are scoring a little more often than they might.
As far as I can gather, the most commonly used xG is that of Wyscout, and I believe it is their numbers being referenced by Parkinson. Wyscout’s website states that they generate xG from the following:
- Location of the shot
- Location of the assist
- Foot or head
- Assist type
- Was there a dribble of a field player or a goalkeeper immediately before the shot?
- Is it coming from a set piece?
- Was the shot a counterattack or did it happen in a transition?
- Tagger’s assessment of the danger of the shot
The last bullet is key. There is an element of ‘how tricky was the chance to finish’, albeit subjective, built into the numbers. What we have uncovered above suggests the assessment could be improved, however.
But there is another factor I have intentionally ignored up to this point. Penalty kicks!
For each penalty, the side in question is awarded a goal probability (xG) of 0.76. The four sides we have looked at have therefore picked up the following values of xGf from just penalties this season:
- Lincoln: 4.56 xGf from six penalties.
- Peterborough: 1.52 xGf from two penalties.
- Portsmouth: 2.28 xGf from three penalties.
- Sunderland: 3.04 xGf from four penalties.
If these penalty kick generated xGf values are removed from consideration, the impact on xGf per game for the season is as follows:
- Lincoln: 1.76 xGf becomes 1.35 xGf.
- Peterborough: 1.59 xGf becomes 1.46 xGf.
- Portsmouth: 1.63 xGf becomes 1.42 xGf.
- Sunderland: 1.83 xGf becomes 1.55 xGf.
Sunderland still come out on top, even with penalties taken out of consideration.
Whatever xGf says (and I am increasingly losing faith in it as a tool), Sunderland are getting roughly the expected number of shots on target. And they are also getting about the expected number of goals.
Rather than exclusively being an issue of finishing - although that is undoubtedly a factor - the low shots on target and resultant goal returns are more a product of the poor quality of the chances being created.
If Sunderland could find a way to increase the frequency with which the attacking players have a clear path to the goal, this season could look very different. But to achieve such a thing would require a different approach that called upon a combination of pace, dribbling skill and intelligent passing.
As things stand, the chances are falling to largely isolated and often static forward players who are impeded by an average of 2.20 players between them and the goalkeeper when the shot is attempted.
It’s hard to see how this can be improved given the relatively low frequency of our attacks (compared to Portsmouth) and the difficulty of the chances being created (compared to Peterborough and Lincoln).
A rethink is needed.