Swansea’s next season will be eagerly watched by the stats community, though not with the hipster adoration with which they will watch, say, Brentford, but rather a more vindictive expectation of statistical schadenfreude. In case you’ve somehow missed the news, Swansea’s ‘underlying numbers’ last season were depressing.

I covered most of the retrospective analysis in a surprisingly popular piece here on StatsBomb. The piece was actually referred to in the Racing Post’s Season Preview pull-out, as can be seen below. I’ll try to not just repeat what I said in there, but look at a few other things about the Swans’ upcoming season.

Racing Post

 

Repeatability

Much of my last piece focused on repeatability – broadly speaking, the repeatability of ExpG over-performance relies on the repeatability of the particular things that were fueling it. For Swansea last season, that was FabiansKI: Fab’s unreal danger-zone save percentage and Ki’s ridiculous conversion rate.

But what if Swansea had a history of Expected Goals over-performance? This is something I couldn’t really look into until Paul Riley released 5 years of team-level Expected Goals data recently, so a huge thanks to him.

Using [Expected Goal-Difference – Actual Goal Difference] as a measure for ExpG over-performance, something very interesting showed up:

Swansea1

 

Swansea out-performed ExpGD by more than 10 goals in 3 of their last 4 seasons. 

This would be worth little, in itself, if ExpGD wasn’t as good of an indicator of GD as it is:

Swansea2

 

And, given that there’s a standard error of around 6 goals either way over that time, Swansea’s consistent over-performance is significant in the sense that it drastically increases the likelihood that, rather than a hugely lucky one-off, their 2014-15 season might have been another of their seasons in which they out-perform statistical metrics.

Again, we’re talking in likelihoods. What would bolster this claim is if Swansea’s repeated over-performance was fueled by the same thing every year, which would show up in their type of over-performance. If we separate the performances (relative to ExpG) of their attack and defence, we can see what has been fuelling Swansea’s over-performance:

Swansea3

 

Apart from a bumper attack in the 13/14 season that off-set huge defensive under-performance, Swansea’s defence has been the main driving force of their over-performance. As each one of Swansea’s seasons in which they over-perform came as a result of over-performing more in defence than attack, it remains increasingly likely that Swansea may be doing something that the model can’t pick up. Not certain, or even probable, but ‘more likely’.

What makes this a bit more confusing is the change from Laudrup-Monk during the 13/14 season, perhaps a factor in that season’s anomalistic stats. Anywho, enough on this I think – I’m swiftly surpassing my own statistical abilities. If someone else has looked into ExpG over-performance (I know Dustin Ward did with Gladbach) and has any opinions, do let me know.

But, in a minor conclusion here, it may be less of a simple case of luck than even I made it out to be. Expected Goals Models find it harder to predict defence than attack, generally, and they may have a problem analysing what could be an idiosyncratic Swansea defence. Did I phrase that carefully enough? You get the picture. The over-performance, then, might be repeatable (he says, gleefully, but apprehensively aware of the ever-looming confirmation bias problems here).

The Player Level

Last Season’s Stars: Attack

Arguably the biggest success of Swansea’s off-season was holding onto their star players. Their real strength is in the midfield that Monk began to cater to in the second half of the season with a midfield diamond. – Jack Cork (signed in January), Ki Sung-yueng, Jonjo Shelvey and Gylfi Sigurdsson.

Gylfi is the main attacking force, managing 10 assists and 7 goals in the league last season.

Swansea4

His P90 numbers were pretty good, though they would have been better had his form not dipped in the second half of the season. The general fan narrative has been to blame the switch to a diamond for Gylfi’s drop in form, but I actually think the diamond suits him. His drop in form, to me, could have as much been about losing Wilfried Bony, who he’d developed a devastating connection with, halfway through the season. On top of that, Gylfi was a bit-part player at Spurs and may have naturally tired out.

Although he behaved pretty poorly when Bony was in front of him, Bafetimbi Gomis came into his own at the end of the season, scoring 5 in his last 6 appearances. His 0.41 ExpG P90 is above average but not quite elite, while his actual NPG90 of 0.38 implies that although he may not be finishing everything, his actual output is roughly in line with what is expected. I’m quietly optimistic about Gomis this season – I don’t see any reason why he couldn’t score 15 league goals if he stays fit.

Last Season’s Stars: Defence

An interesting stat, that. But not really worth anything in itself. Swansea will want Williams to keep up his form from last season, though I’d wager metrics that prioritise on-the-ball actions less like Shapley values would rate Williams higher than looking at his raw output (I could be completely wrong there). Last season’s signing Fede Fernandez settled in well bar a horrific showing against Chelsea, which bodes well for the first game of the season. Kyle Naughton was surprisingly good before being injured, and will surely hold onto a starting spot over Angel Rangel when fit.

Obviously, the real hero of last season was Fabianski. I wouldn’t bet on him keeping up his obscenely high danger-zone save percentage, but he is still undoubtedly better than the general impression of him before he moved to the Liberty.

Recruits

Andre Ayew. What a signing – the player I was always going to waste my radar wish on:

Ayew Radar

 

Ayew has been described as a ‘coup’ a lot, but considering the general level of player acquisition in the Premier League’s midtable, I’d be wary of calling him that. He is a very good player, though, and arrives off the back of a good season in the madness of a Bielsa team. He interestingly has a high defensive output for a winger and Monk will like this. Where he will play depends on the formation: left-wing in a 4231, just off Gomis in a 41212.

Swansea also signed Eder, Franck Tabanou and Kristoffer Nordfeldt. Tabanou is the only one with a real possibility of breaking into the first team over Neil Taylor, while the other two will likely deputise for Gomis and Fabianski respectively. These squad signings mean that Swansea’s 25-man arsenal will be as strong as it as ever been in the league.

Prediction

Anywhere from 8th-14th, really, depending on whether or not any players have a great season. It’s hard to see Swansea getting dragged into a relegation battle, but it may be a tad optimistic to hope for another record breaking campaign. Monk summed up his own plan for the season in a typically pragmatic fashion: ‘Get to 40 points as quickly as possible, and push on’.

 

 

I’m on Twitter @BobbyGardiner. A big thanks to Paul Riley, Per Linde and @stats4footy whose data I used.

 

  • Birdace

    Concerning Gomis, as a Lyon’s supporter, he is the typical average striker and one of the most consistent players in my data in terms of ExpGoals. He will always score you between 10 and 15 goals and he is never injured. But, in the other hand, he isn’t really creative, can’t dribble and loses the ball too much for my taste.

    Ayew is a really good player, very versatile and he like to play in central midfield where he played well. Tabanou is a former winger and a good player too. Nice recrutment by Swansea, in my opinion.

Improve Performance and Productivity in Your Club:
State-of-the-art Football Analytics