It’s been a confusing season to be both a Swansea fan and maintain an interest in analytics. On one hand, a record points finish and altogether hugely celebrated season; but on the other, repeated pessimism about Swansea’s “underlying numbers” from the analytics community:
I’ve been left in a sort of un-opinionated grey area – I know that mere point totals aren’t everything, but I’m similarly aware of problems with metrics in general. A full season since I claimed to be optimistic about Swansea’s chances of breaking into the top half in my StatsBomb Season Preview, I’m unable to gloat shamelessly.
And so, people of the analytics underworld, Swansea fans and any neutrals brave or interested enough to read this, I shall attempt to answer the question that undoubtedly keeps you up at night – “Are Swansea crap?”
Why would they be?
Swansea didn’t shoot much this season relative to how many shots they conceded, basically. Their Total Shots Ratio (TSR) is the 3rd lowest in the league, better than only Burnley and Sunderland. TSR is a reasonably good indicator of team performance:
But it isn’t perfect. Swansea are obviously anomalistic here, but two other teams are similarly so: Chelsea and QPR, who over and underperformed respectively based on their TSR. There’s some explanatory power lost in TSR – this season, the power to explain the best and the worst teams in the league.
Which is why the analytics community upgraded with Expected Goals. In layman’s terms, ExpG are the amount of goals a team might have expected to score/concede in a game based on historical trends, predominantly location i.e. Bobby shot from X yards away, and he might have expected a Z chance of scoring. In more complicated terms, they’re this.
As can be seen immediately, ExpG Ratio (ratio of ExpG for and against) was markedly better at explaining point performance this season than TSR. Chelsea and QPR still over/underperformed, but by less. Swansea, though, are still a prominent outlier, their 16th placed ExpGR contrasting harshly with their 8th placed finish.
And this is where one might conclude Swansea are ‘crap’ (or at least not good enough to be 8th) and lucky. Not one of the two, but both. Their point performance should, over some period of time, regress, possibly deflating the wheels of the Monk bandwagon balloon before it had any time to properly get going.
Picking apart their overperformance
Michael Caley’s wonderfully public ExpG model has Swansea scoring 41.2 goals last year and conceding 53.7. In reality, they scored 46 and conceded 49, which (+4.8 and -4.7) is almost identical overperformance in both Swansea’s attacking and defending.
The question, then, becomes not “Are Swansea crap?” but “Is there anything that might suggest Swansea’s overperformance is a product of anything other than luck/variance?” A less catchy title, and one with huge confirmation bias problems for a Swansea fan to examine. But maybe, just maybe, Swansea are anomalistic because of footballing idiosyncrasies.
Unusual attacking style
Swansea are a hipster club for a reason, their possession football mantra setting them apart from the otherwise boring mediocrity of mid-table. This season, though, Swansea’s football has been (on the surface) much less atypical. Monk is far more focused on adaptability and contextualised match tactics rather than the poetic Laudrupian ‘go forth and pass’ mantra, which has been hard for people to criticise given the record points total etc.
But if you look at the type of shots Swansea take, they remain unusual for a mid-table side. The percentage of shots they take from a through-ball (6.4%) is the fifth highest in the league, with Arsenal, Manchester United, West Brom and Manchester City ahead of them in this particular measure. And the percentage of shots they take from a cross (40.7%) is the sixth lowest; only Chelsea, Liverpool, Arsenal, City and United rely on crossing less. Swansea’s attack shares some of the characteristics with the ‘big sides’ that tend to differentiate them from the minnows, but differs in terms of volume of production.
Examining United’s overperformance in the 2012/13 season, Daniel Altman highlighted two possibilities (other than luck) – set piece goals, and Robin Van Persie. Swansea, however, scored the least set piece goals last season (4) and Gomis, although he came to late form, only scored 7 of his 69 Premier League shots while Bony, at the club until January, scored 9 of his 66.
Ki Sung Yueng managed 8 goals from 28 shots though, giving him a ridiculous conversion rate of 28.6% from centre-midfield. According to objective football’s numbers, this is the highest shot conversion rate of ANYONE in the league to have played more than 1000 minutes. I’m trying desperately to avoid a “the key to…is Ki” pun here, for the record, but Monk’s switch to a diamond in January allowed Ki more freedom to attack and he definitely made the most of it.
It’s unlikely that Swansea’s over-performance in attack comes from them creating chances uniquely unexplainable by ExpG, especially given their similarity in style with the big clubs for whom ExpG was pretty accurate this season. A fair portion of their 4.8 goal over-performance might be explained by Ki’s unusually high conversion rate. Sadly for us Swans, this isn’t particularly repeatable – Ki’s own conversion rates were 7.5% and 0% in the two seasons prior to this one.
For Swansea’s attack to score as many or more next season, Monk should look to drastically increase the amount of passes they make in the final third. Only Palace manage less than Swansea per90, and that suits their extremely different attacking style more. Gylfi Sigurdsson is key to this. His drop in performance in the second half of the season may have been because of a formation change, but I’d be wary of viewing Bony’s exit as only a loss of goals. Bony’s hold up play facilitated Swansea’s slower attacks and gave Gylfi freedom, Bafe Gomis has been less good at this despite his late surge of goals.
PDO is a loose measure for ‘luck’ or unsustainable variance. It combines a shooting component (goals for/shots on target for) and a saves component (100 minus goals against/shots on target against). Swansea’s PDO for last season was 105.4, the 4th highest in the league and noticeably higher than the league average of 100.
Based on what we’ve already been through, it may be obvious that Swansea’s high PDO isn’t primarily attack driven: Swansea’s shooting component (scoring %) is 31.4, closer to the league average of 30.4 than any other team.
One man is the reason for Swansea’s high PDO, and his name is Lukasz Fabianski. Swansea’s saves component (save %) is 74.3, the highest in the league, followed closely by Chelsea with 74.2 and less closely by West Ham with 73.7. According to Paul Riley, Fabianski has the highest danger zone shots saved percentage of any goalkeeper in the league in the last five years. This would go some way to explaining Swansea’s over-performance in defence against ExpG, which can’t account for Fabianski’s ridiculous shot stopping. In one of his weekly round-ups, James Yorke illustrated the effects of a noticeable difference in save percentage on the output of the otherwise statistically similar Swansea and Leicester.
So that’s good, right – Swansea have a great keeper? Well, much like with Ki but to a far more important degree, the issue here is repeatability. Fabianski might be a fantastic keeper, and is almost certainly better than the general perception of him a year ago, but keeping up this level of shot-stopping is extremely unlikely. Swansea need to look at stopping the danger zone shots altogether, not relying on Fabianski to save them.
Wrapping it up
Are Swansea crap? Err. Umm. Completely objectively, they’re probably lucky to be 8th, especially given they’re the 6th worst team both in terms of ExpG for and against. Through separating and cross examining attack and defence, we’ve seen that the factors that probably pushed their ExpG overperformance to allow them to finish so high are also unlikely to be repeated. This isn’t great news for a Swansea fan.
But issues only really start to arise if Swansea, and Monk in particular, were to take their 8th placed finish for granted. Of course luck is involved, albeit to an arguable degree; it would be criminal for Monk to look back at the 1-0 away wins at Arsenal, United and Southampton and not realise that in most timelines you come back with 0 points, not 9.
My analysis has been inherently retrospective, but how well one can evaluate this season depends hugely on the next. Say Swansea post similar numbers and still finish 8th, it becomes increasingly likely that they’re doing something unique (and repeatable) that the models aren’t picking up. Sure, hoping this happens might be naïve on my part, but isn’t that kind of the point of supporting a football team?
I’d be really interested in any thoughts on this piece/methodology.
You can find me at @BobbyGardiner on the ol’ Twitter.