The Stat Man: Sunderland’s shots on target problem and what it means for promotion

If you’ve read some of my previous articles, follow me on Twitter, or are unfortunate enough to be stuck in a group chat with me, you will know I talk a lot about shots on target and how important they are.

In order to judge whether I consider a performance to be ‘good’, or not, I ask two questions. The first is, did we get five or more shots on target? The second is did the opposition get fewer than four shots on target? If the answer to either question is no, then, according to my admittedly binary criteria, the performance was not good. Conversely, if the answer to both questions is yes, then I’m pleased that we put in a good performance.

This overly simplistic point of view reduces ‘the beautiful game’ to a couple of numbers. However, contrary to the impression I might give, when I go to the match, I’m not keeping a tally of shots on target and using that information to determine whether I’m having a good time or not! Even stats nerds enjoy a few pints and a good shout at the match.

While the shots on target criteria may be simplistic, it is not arbitrary. Allow me to explain.

Below is a similar chart to one I have used on a regular basis to demonstrate a team’s attacking quality. Usually it contains all the League One teams in the current season, but for this explanation I have done away with everyone except our beloved Sunderland. I have also included the last six teams to be automatically promoted from League One. The chart is correct as of February 6th, 2020.

The diagonal dotted line represents the mean proportion of shots on target that result in a goal in League One. You can think of it as ‘expected goals’ but not the same expected goals as the much more complex number you generally see presented as ‘xG’. Using this line is simple, look at the number of shots on target on the x-axis, then see the corresponding expected number of goals on the y-axis. It’s just y = 0.33x (i.e. expected goals = 0.33 x shots on target). Nothing fancy.

Something five of the six automatically promoted sides have in common is having an average of five or more shots on target across all games. Getting five shots on target, based on the expected goals formula, results in 0.33 x 5 = 1.65 goals. Goals only come in integers of course, and in those terms, 1.65 could be thought of as a team being very likely to score at least one goal and also to score more than one goal more often than not.

Taking Luton in 2018/19 as a specific example, we find they averaged 5.96 shots on target and this translated to an average goals per game return of 1.96. Pretty impressive. Translated into real world goals, this means Luton would almost certainly score at least one goal, and score more than one goal very often. I checked their results from last season and found they scored one or more goals in 89% of games and two or more in 65%. So not a bad approximation.

Examining Luton’s position on the attacking plot, you can see that their datapoint is almost exactly on the dotted diagonal line. This means they came close to ‘achieving’ the League One average of scoring from 33% of their shots on target. A quick check of the maths confirms this (1.96 / 5.96 = 0.329).

Now look at Sheffield United in 16/17 and Barnsley in 18/19. I note the following:

1. Neither team are on the dotted line.
2. Barnsley averaged more shots on target than Sheffield United.
3. Sheffield United scored more goals than Barnsley.

The position of the two teams relative to the line tells us about the quality of their finishing. Sheffield United ‘only’ averaged 5.43 shots on target but scored an average of exactly 2.00. Barnsley, on the other hand, generated an average of 5.65 shots on target yet ‘only’ scored an average of 1.74 goals per game. These results indicate that Sheffield United’s finishing was above average, hence their datapoint is above the line, and Barnsley’s finishing was below average, resulting in their datapoint being below the line.

Sheffield United’s result tells us that it is possible to create fewer clear goal scoring chances than another team yet still score more. But there is a limit to how far ‘above the line’ a team can stray, however good their finishing is. It would be extremely unlikely that a side could, for example, generate an average of five shots on target per game and score an average of three goals. That type of thing does happen in occasional games of course. A clear example of this can be found in Sunderland’s home game against Tranmere Rovers this season when we generated six shots on target and scored from five of them. But if a return of six shots on target were achieved in ten consecutive games, there is no way five goals will be scored every time.

Just as a team generating five shots on target, on average, will not score three goals per game, on average, the same fundamental limitation applies to teams only generating, say, 3.61 shots on target, such as Sunderland this season. If it’s very difficult to average three goals per game from five shots on target per game, the same applies to trying to average two goals per game from 3.61 shots on target. It’s not going to happen. and this is reflected in Sunderland’s current goals per game average of 1.32.

Referring back to the attacking plot, the six automatically promoted sides are clustered a) to the right of the attacking plot and, b) around the dotted line. Sunderland - while happily positioned above the line, indicating there is some above average finishing at play - are much too far to the left to be automatic promotion challengers. Even the poorest of the promoted teams (Phil Parkinson’s Bolton in 16/17) were much more effective in creating chances.

Sunderland will not be automatically promoted based only on their attacking performance unless something dramatically changes to vastly increase the number of shots on target from now until the end of the season.

What of the defensive side of the game? It has been a feature of our play that, since Parkinson’s arrival, Sunderland have developed the best defence in League One. Perhaps a brilliant defence is enough to cover the attacking shortcomings? Let’s examine that.

Here we see the same chart components but this time from the defensive angle. Again, the last six automatically promoted teams are plotted along with Sunderland this season and a line showing the League One average. This time, the datapoints are plotted with shots on target ‘given way’ along the x-axis and goals conceded on the y-axis.

Interestingly, all teams, including Sunderland, are below the League One average line. This means all have goalkeepers who are outperforming the league average for the proportion of shots saved.

The standout performance comes from Wigan Athletic, who gave away an average of 2.26 shots on target per game and conceded an average of 0.63 goals. The rest of the performances - including Sunderland - are all very similar (if you’re struggling to find us, we’re experiencing a Luton-eclipse in the plot).

Ignoring a freakishly good Wigan, this season’s Sunderland are right in among the success stories from the last three seasons. Our defensive performance is certainly good enough to gain automatic promotion.

At this stage, I will refer you back to something I stated earlier in the article.

In order to judge whether I consider a performance to be ‘good’, or not, I ask two questions. The first is, did we get five or more shots on target? The second is did the opposition get fewer than four shots on target? If the answer to either question is no, then, according to my admittedly binary criteria, the performance was not good. Conversely, if the answer to both questions is yes, then I’m pleased that we put in a good performance.

Now I have presented the data, I hope you understand my attitude a little better. Automatically promoted teams tend to get five or more shots on target and give away less than four.

So far, we have learned that our defence is up to scratch but our attack certainly isn’t. But could having the top defence in the league perhaps be enough to get us over the line to automatic promotion?

To begin to attempt to answer this question, it is necessary to combine the attacking and defensive numbers somehow. And this is where things take a slight step in the abstract direction, but stick with me because it’s still pretty simple.

By dividing the shots on target ‘against’ by the shot on target ‘for’, you obtain a ‘shots on target ratio’. Similarly, by dividing the goals ‘against’ by the goals ‘for’, you obtain a goals ratio. Both are plotted below.

The closer a datapoint is to the bottom left corner, the better the overall performance. As implied by the arrow on the plot, perfection is achieved when both the shots on target ratio and the goals ratio are zero. Perfection is near impossible however. It would require that a team don’t give away any shots on target to the opposition or concede any goals for the entire season! Wouldn’t it be nice to be that chilled out as a fan?

Wigan Athletic - who were not the top attacking performers but were excellent in defence - lead the field for overall performance, occupying the position closest to ‘perfection’.

Sunderland, unfortunately, are off the pace. As good as the defence is this season, the attacking shortcomings are too detrimental to overall performance for automatic promotion to be likely. Even the ‘worst’ of the automatically promoted teams - Bolton - are well ahead of us.

The results are fairly clear from this overall performance chart, but the data takes into account both shots on target and goals. My original conjecture was that shots on target are the most important aspect of performance. So let’s remove goals from the equation and only examine shots on target.

The data has now been reduced to a single dimension so is best presented in a table.

The results are not a surprise given what we have already learned about the previously automatically promoted teams compared to Sunderland in 19/20. But it is encouraging that, even after removing goals from the equation, the order from best to worst remains unchanged.

Using the same method of only looking at shots on target, here is what the current League One table would look like (left) compared to the current actual League One table based on points per game played (right).

Some things are not surprising, for example, Bolton and Southend are in the bottom three of both tables. The top two in the shots on target table could be considered a little unexpected, however. Portsmouth are top, and where did Oxford come from?

Looking at Sunderland’s position is discouraging. Based on the theoretical importance of shots on target, Sunderland might finish 10th and Oxford might finish 2nd. Ouch!

I don’t know if that is what will happen. If I could be that accurate about the future I would be rich and I probably wouldn’t be spending my time telling you about my nerdy approaches to football data (or maybe I would be doing it while sipping a cocktail next to a pool). What I am able to do though, is look at an example from the 2018/19 League One season taken in early October.

Compare the tables on the left and right. Not perfect, but not too shabby either. Particularly when you consider that Barnsley were 4th and Luton were 12th in League One at this time. Three of the four play off teams are correctly predicted too.

Will Sunderland finish in and around 10th place this season? Will Oxford finish second behind Portsmouth? I don’t know. Things can change. Portsmouth are certainly on a roll but Oxford have lost two very important players in the transfer window. Sunderland have brought in a set of new players that might - when combined with Phil Parkinson’s experience - make a difference.

Whilst I cannot predict how the rest of the season will play out with certainty, I can say that, unless Sunderland start to create a lot more chances, the probability of gaining automatic promotion is close to zero.

Let’s see what you’re made of Phil.