Arsenal’s Back Three Myth Busted: By the Numbers


There were two distinct periods of play for Arsenal this season: the first 30 League games where Wenger played with a back four and the final 10 game stretch of the season, which includes two FA Cup matches, where Wenger played with a back three (five). Here is a chart with the data:

The first thing that stands out is the remarkable similarities between shots taken (14.9 v. 14.4) goals scored (1.9), Shots in Prime allowed (3.2 v. 3.4) and Big Chances allowed (1.5). For those who don’t know, Shots in Prime are shots taken within 11m of the goal line in a hemi-circle from the penalty spot and Big Chances are shots taken from extreme close range or with very few defenders (usually 1v1) in front of the goal.

A note on methodology: I compile all of my own shots stats. I use the 442 Stats Zone App after every Premier League game to map where the shot was taken from and whether Opta marked it as a “big chance” or not. Using that data, collected now over three years, I find that Big Chances are the most important shot in the game. They are scored at a 50% rate and they account for 50% of a team’s goals. Shots in Prime are not really correlated as well to goalscoring and once I remove the Big Chances I find that they score at about a 10% rate. This is very close to the bulk average for all shots across the pitch. Shots from outside the 18 yard box are the most numerous type of shot and the least effective, scoring at a 3% rate. Now, back to the chart.

Where there are differences are in the attack. Arsenal created 0.8 more Big Chances per game with the back three than they did with the back four. This leads to an increase in Shots in Prime as well, since those two numbers overlap. These numbers lead to an increase in expected goals per game from 1.9 to 2.2. That doesn’t seem like much but over a 38 game season that’s 11 goals and would probably have been enough of a difference to ensure a top four finish. So, the back three, which is typically thought of as a defensive formation, actually led to a renaissance in Arsenal’s attack.

This also shows up in Alexis’ goal-scoring and assists. Alexis scored or assisted 9 total goals in the final 10 matches. In the previous 10 matches that number was down to 6 and included a stretch of 6 matches with just two goals.

Further, when I compare ONLY the matches contested between similar opponents (Chelsea, Man City, Man U, Leicester, Middlesbrough, Stoke, Sunderland, Everton, Tottenham, and Southampton) we see once again almost no change in the defensive stats but a significant improvement in attack. Even more improvement in attack than against all opponents.

Arsenal created 1.3 MORE Big Chances with a back three against the same opponents than they did with a back four. And the resulting better shots selection increased Arsenal’s expected goals for by 0.7 per game. Those are remarkable numbers and I feel comfortable saying that it looks like the back three significantly improved getting the ball forward.

If I’m allowed a bit of pontification, I’d suggest this was down to putting Ramsey in the middle. Ramsey is a much more attack-minded midfielder than Coquelin or Elneny and while his runs into the opposition box make some fans crazy (me) this evidence suggests that they may improve Arsenal’s attack.

But where the back three didn’t make any improvement was in Arsenal’s defense. This is exactly contrary to what most fans have expressed in the forums, on twitter, and to me personally. Now, before I go any further, you have to be consistent here. You have to either accept the shots data both for and against Arsenal or not accept this data at all. It would be inconsistent to say that Arsenal’s attack improved but then discount the same data to say that Arsenal’s defense also improved.

Similarly, it would be inconsistent to cry “small sample size” to argue against my analysis of the back three while also saying you think the back three worked well. The “small sample size” argument cuts both directions. I don’t believe that 25% (10) of a 40 match set is a small sample so I am comfortable with my conclusions.

Back to the defense. There are a number of problematic stats here. This first is just total number of shots. Comparing like for like, Arsenal allowed 2.8 more shots per game with a back three than a back four. The number of Big Chances allowed per game stayed steady but Shots in Prime went up marginally. As a result, I expected to see more goals allowed, about twice as many.

“Ah ha! But we DIDN’T! This means the back three worked!” you might say. Ehhh, no. Instead, what I saw was a massive increase in the shots saved by Cech and Ospina – an increase over both the 30 game larger sample and the like-for-like sample. Those 46% Big Chance saves and 69% SiP saves are astonishingly high. Hugo Lloris was the best keeper among the top six teams, he saved just 39% of the Big Chances he faced. The average across the top six teams was just 31% with David de Gea saving just 20%. And the highest SiP saves was Man City with 56% and Liverpool and Man U saving just 33%.

Using the shots faced data and my own expected goals formula I gave Arsenal’s keeper either a + or – rating on each of the 40 games I sampled this season. +1 means that the keeper saved 1 expected goal, -1 means the keeper failed an expected save.

This chart shows a remarkable increase in Arsenal’s goalkeeping in the final 10 matches of the season. Not quite as good as the first 10 matches, but consistently good at the end of the season with no negatives.

What I would expect to see from an improved defense would be a reduction in the number of shots, the number of big chances, the number of shots allowed in prime, and thus a reduction in the expected goals allowed. What I see instead is simply an improvement, an unsustainable improvement, in Arsenal’s saves.

One final statistical note. Arsene Wenger is a huge proponent of scoring first and scoring early. He has said many times that the team who scores first goes on to win most of the time (and he’s used different percentages, but they are all above 70%). I can’t confirm this stat but what I can confirm is that in the first 30 League games either Arsenal scored or conceded on average in the 29th minute. But when Arsenal switched to the back four, three¬†the first goal scored or conceded drastically changed and moved to the 49th minute on average. This is quite remarkable and there could be a number of explanations that exclude the back four/back three debate – late season fatigue or dropping Mustafi are two examples. But it is true that in the 10 game set directly prior to Wenger changing to a back three, Arsenal conceded the first goal 6 times and did so on average in the 11th minute. And Arsenal lost all 6 of those matches.

It doesn’t look like the back three made Wenger’s Arsenal more cautious, their attack remained steady. Nor does it look like the back three made Arsenal more solid in terms of shot concession. But it does look like there may have been a psychological boost to the team, which when coupled with an increase in Keeper prowess cut out the early mistakes and gave Arsenal some room to breathe life into a dead season.


Stats: my personal database


Leave a Reply

109 Comments on "Arsenal’s Back Three Myth Busted: By the Numbers"

Sort by:   newest | oldest | most voted