This is going to be a
really long post - its probably more of an 5-10 min 'article style' read than a 20 sec post, so bear that in mind. If you have made it beyond that sentence, then prepare to strap in, and get ready to feast on some juicy analysis. Or just ignore this post entirely and go back to moaning about how all statistics are useless or how it’s all Stams/Aluko’s/Brian’s fault. I don’t mind.
While I accept the view that we should just let this season die and never speak of it again, I think there are some really interesting stuff borne out of the statistics (I have used WhoScored.com for all figures, although all the calculations/errors are my own). Personally, I see a lot of good intention and effort went into the OP, but I think we need to take a step back and think more holistically. There are certainly more pieces to this jigsaw puzzle, which have all been talked about in this thread and elsewhere (e.g. lack of fitness, player attitudes, confidence, tactics), but I think this post will add evidence that it was the injuries and transfers that hurt us most.
This season we conceded 70 goals while last season we conceded 64 goals – not hugely different, although a slight underperformance (maybe one to explore later…). We might have foreseen this after selling AAH and had we known McShane would miss a large part of the season. However, for me the stand out change was the sharp fall in goals scored – from 68 last season to only 48 this season. It won’t come as a surprise that there is a decent statistical relationship between goals scored and points amassed across the whole league – the correlation is 0.72 this season (down from 0.83 last season). People will rightly point out that the correlation for Goal Difference (0.95(!) this season, 0.77 last season) and Goals Conceded (-0.95(!) this season, -0.75 last season) are empirically stronger/more important, but since the main change in was our attacking output, I will focus on that.
I’ve plotted this relationship (goals scored vs points won) for the whole championship over the last 2 seasons, along with a simple average trendline for the relationship. Anyone sat above the trendline has performed ‘better’ (amassed more points than their total number of goals alone would suggest), anyone below has performed worse. You can see we went from outperforming this very simple relationship to underperforming. Put another (extremely crude) way, if this relationship were perfect then last season our 68 goals
should have been worth 13 points fewer than it was (placing us 8th), and this season our 48 goals
should have been worth 7 points more (up to 18th). In reality this is a very extreme extrapolation of a simple relationship, so don’t read too much into it, although it does support the popular ‘false third’ argument and frequent narrow victories last season, as well as some bad luck this season.
The reason this is more relevant for us relates to the breakdown of results across the two seasons. In 2016/17 our most modal or most common winning margin was 1 goal (18 occurrences), while this season it has been 0 (game drawn – 14 occurrences). In a sport of such small margins, where we were only winning by those same small margin, a slight change can have a big effect. In this case the slight change was our attacking output, where lots of 3 point 1 goal wins switched to 1 point draws.
For this piece I have analysed some basic statistics on the attacking unit compared to last year.
For the conclusion up front, a series of downgrades (via injuries and transfers) across all attacking position groups cost us badly this season in aggregate.For our ‘attacking unit’ season comparison I have taken 1x striker, 2x wingers, 3x (creative) CMs and 2x FBs, so basically everyone who isn’t a CB – the composition of the units are outlined in the table below, and are highly subjective. I appreciate playing all three CMs at once is unrealistic, but it allows a direct comparison of the more attack minded midfielders (rather than Van den Berg or Evans), and generally 2 of 3 played most games in both seasons, so a 3 v 3 comparison ‘feels’ fair at least – the main criticism is it will inflate the attacking output (as we only played 2/3 of the CMs at any given time), but this critique is applicable to both seasons, and I wanted to capture all of these players performance. If you want, feel free to deflate all the midfield stats by 1/3 so they represent 2 ‘whole’ players, more in line with our line ups.
Where two players are listed, I have taken the weighted average (by minutes played) to construct a 2 player ‘hybrid’ purely to allow a fair like forlike comparisons across seasons. Where a player has played multiple positions (e.g. Bacuna RB/CM and Blackett LB/CB) I have only used games where they have played the relevant position.
On the face of it, the composition of the ‘attacking unit’ doesn’t look that much worse, wingers should be comparable and Edwards would be considered more attacking than Danny Williams. The obvious big downgrades here are Kermorgant to Kermorgant/JDB, and Obita to Richards/Blackett.
Nevertheless, every single position group underperformed this season, at almost every metric I looked at. It’s worth noting here that our average possession over the season was pretty much unchanged (53.8% this season, from 55% last season), so it’s just
what we did with it that differed. Shots per gameOver the whole unit, the average number of shots per 90 mins fell from 12.3 to 9.7, with 17-27% falls by CMs, Wingers and Strikers. Despite Barrow scoring a lot, he actually only averaged 1.3 shots per 90, compared with Beerans’ 1.5 per 90 last season, while on the other wing Aluko’s 1.7 shots per 90 were lower than GMac’s 2016/17 2.1 per game. In the middle, Liam Kelly actually shot slightly more frequently (1.3 per 90 vs 1.2 last season). The falls came from Edwards 0.7 shots per game (vs 1.5 in his final season with Wolves) vs Danny’s 1.6, and Swift/Bacuna hybrids 1.6 vs Swift’s 2.1 last season (hybrid split – JS 1.9 this season, LB 1.2).
In aggregate, this is large difference. Last season 11.8% of our shots resulted in a goal, this season only slightly less at 10.5%, but the attacking unit taking 2.6 fewer shots as a whole per game translates to around 0.28 fewer goals per game (very crudely approx. 0.3 points per game). That’s a lot over a season. Despite falls across the board, it’s the reduced output from CMs that stands out to me – on the wings the loss of GMac’s 2016/17 season goals (albeit frequently from the penalty spot), has been mostly compensated by Barrows 2017/18 goals.
However, I don’t think you can solely place the blame on a few players being ‘shot shy’ here, since to have a shot first you must make a chance. I logically assert that the two main sources of shots (from open play) will be through the centre mids, and down the channels (wingers and full/wing backs).
1 - Middle - Key passes (defined as passes resulting in a shot – it’s unclear if WhoScored.com count only unsuccessful chances, or do include assists in this measure)
Key passes per 90 mins fell across all positions, with total Key Passes per game falling by a quarter- from 10.1 to 7.4 per game. The major fall was Key Passes from wingers, with Barrow surprisingly only offering 1.0 key passes per 90mins compared with Beerans’ 1.4 last season, and Aluko making 1.2 (the same as in his previous season with Fulham) compared to GMac’s 2.3 in 2016/17. In CM, Kelly actually performed slightly
better this season (1.8 vs 1.7), but this was more than offset by Edwards only offered 0.4 to Danny Williams’ 0.6 last season, despite the former being billed as a more attacking replacement. As a side note, Edwards Key passes per 90 minutes were significantly lower than over the rest of his career, and half the 0.8 per 90 that he registered last season at Wolves. Finally, completing the CM area, there was a big fall from 2016/17 Swift’s 1.5 key passes, to the Swift/Bacuna hybrid of 0.89 (although a crocked Swift himself averaged 1.3 per 90 this season, when he wasn’t on the physio’s table).
Verdict – Downgrades on both wings (two transfers in, one transfer out one bad injury), while another costly significant injury to our most offensive CM in John Swift.
2 - Channels - crossesCompared with last season we attempted 43% fewer crosses, a huge fall especially considering we had broadly the same amount of possession. However, the share of attempted crosses that were ‘accurate’ (the definition is unclear, let’s assume it means either met by a teammate, or crossed into a somehow defined ‘dangerous area’), is broadly unchanged season on season (at 21% vs 23%), and doesn’t materially differ between Fullbacks and Wingers or across seasons. The largest difference by position was a 2016/17 25% cross accuracy from fullbacks, dropping to 20% this season. My initial hypothesis was that this could be an Obita vs Richards/Blackett factor, but in honestly this difference isn’t huge.
Obita averaged 3.4 crosses per game last season (24% of them were accurate), by contrast the Richards/Blackett hybrid only averaged 1.3 (24%). To split them even further, Richards averaged 1.3 crosses per game at a surprisingly poor accuracy of 8% (despite being another converted winger like Obita), while Blackett also averaged 1.3 per game at an equally surprisingly 31% accuracy. This downgrade alone cost us 2 crosses per game, of which 0.5 would be expected to be ‘accurate’, whatever that means.
Gunter’s headline output was reasonably static, but there was a slight increase in frequency (2.3 vs 1.9 attempted) but a big fall in accuracy (0.4 or 17% accurate this season, 0.5 or 26% accurate last season). The fall in accuracy could be due to starting quite a few games at LB and having to more frequently cross with his weaker foot - I chose to include these games as a LB in my sample as they were still genuine games played as a full back.
As for wingers, Aluko’s 4.4 crosses per 90 (1.1 or 25% ‘accurate’) falls below GMac’s 6.7 (1.9 or 28% ‘accurate’) – while this is consistent with the “for o*ford sake Aluko, stop drifting inside” argument, it should be noted that
he crossed the ball 57% more this season than he did last season at Fulham (2.8 per 90, of which 0.6 or 21% accurate). On the other wing Barrow offered 2.9 crosses a game (0.5 or 17% ‘accurate’), well below Beeren’s 6.1(!) per 90 (1.0 or 16% accurate).
Verdict - this season our wingers have seriously underperformed last season, in terms of getting the ball in the box from out wide. This may be due to personnel, or possibly a tactical switch. A drop from 13.9 crosses to 8.6 per game (or 4.2 to 2.3 ‘accurate crosses’ per game) is going to cost you goals in a team spearheaded by a target man…
Player styles – the target man I think the final piece of this puzzle, which isn’t captured above, relates to the individual player styles. You might think JDB is a like-for-like (albeit inferior) replacement for Kermorgant’s target man role, but I think the stats bring out some key differences. I've split this into three sections (i) shots and goals (left/red axis), (ii) Aerial ability (middle section, right/blue axis) and (iii) passing (right/blue axis).
2016/17 Yann was very good. He had more shots, goals, successful aerial duels, passes and accurate passes per 90 minutes than 2017/18's Yann and 2017/18's JDB. But it looks like he is in decline, which should be expected as he approaches 37. When he did play this season, statistically he performed much worse, not just as a striker but also in his link up play. His shots per game fell from 3.1 to 2.5, and his goal per game return fell from an incredible 0.5 to a miserly 0.13 (one in two, to one in eight). Lack of fitness aside, that is a significant drop, and his 1300 odd minutes this season is not a totally insignificant amount. He actually attempted more aerial duels per 90 this season (12.4 vs 10.8 ) but despite this was successful in
fewer (6.2 vs 6.6). I fear this is a growing trend showing his decline, his aerial success rate was only 51%, compared with 61% last year and 68% in his 16 games with us in 2015/16. This 51% success is the worst I can see in his career (at least back to 2010/11), aside from the 42% he managed over just 125mins (7 appearances) with Bournemouth in the PL before joining us. As for link up play, his passes per 90 fell from 30.7 to 27.5, and the number of these which were ‘short accurate passes’ (selected by me as a proxy for linking up with midfielders) plummeted from 22.6 to 13.7. I think 2018/19 expectations for target man Kermorgant should be tapered accordingly, even if he is fully fit and raring to go come August.
So how is JDB as a like for like replacement? Well, aside from being big, I’m not convinced he is a target man, and certainly not capable of leading the line
and linking up play like Yann did in 2016/17. His 0.35 goals per 90 isn’t appalling, especially considering he had only 2.4 shots per game – worth noting that this low figure could be (i) his inability to be in the right place or control the ball before executing a shot, or equally (ii) down to the rest of the team failing to make quality chances for him. Certainly I have shown above how the support from CMs, Wingers and Fullbacks was significantly lower than Yann received in 2016/17. My main concern on JDB is his link up play, winning 4.6 out of 13.4 aerial duels per 90 (34%) doesn’t look good as a candidate to be a meaningful aerial threat – indeed this is in line with the 38% at Wolves the season before, so not a one off. As for passes, 22 per game (13.5 ‘short accurate’ passes) falls well below Yann’s 30.1 (22.6) from 2016/17.
Conclusions (in no particular order):
• Edwards significantly underperformed compared to previous seasons. Maybe he is over the hill, or maybe he is being asked to play a different role – either way his output is around half what we as fans, and whoever made the decision to sign him, would have planned for. I’ve mentioned shots and key passes, but his tackles per game (3.1 to 1.9) and Interceptions (1.7 to 1.5) also fell.
• Aluko appears a very odd signing as a right wing replacement to GMac – it’s possible he wasn’t intended to be like for like replacement (we probably didn’t expect GMac to miss the vast majority of the season), but as a right winger he offered fewer shots, key passes and crosses than GMac did last season. In terms of crosses, he performed better this season than he did in his last season at Fulham, suggesting he simply isn’t suited to the role he played this season/GMac played last season. If he was genuinely signed for that role to compete with GMac then I believe it was a terrible decision to sign him– his history has suggested he doesn’t cross the ball often, and when he does it’s not especially well compared with others. Alongside that, his shots per game have halved compared with last season so perhaps he spent
too much time out wide and should have been instructed to play more in line with his (apparent) strengths…
• Obita is a key offensive full back – Richards hasn't been able to offer the same threat in his breakthrough season (despite a couple of unexpected goals), and while Blackett was seemingly (/surprisingly) dangerous when crossing, he doesn’t get forward as often and the consensus is he is a liability when defending as a LB.
• Getting and keeping John Swift fit is key to increasing our attacking output. While I never felt like he got up to full fitness this season, the only small falls in his offensive output stats suggest it would be too early to write him off entirely.
• Statistically (purely offensively) Liam Kelly doesn’t look as bad as the general consensus would suggest, although I haven’t looked at his passing accuracy which people have commented on quite frequently. Defensively, (not his role imho) his tackles per game did drop compared to last season (2.5 to 1.9), with interceptions unchanged at 1.1
• 2016/17 Yann was very good. Certainly better than 2017/18 Yann, and 2017/18 JDB. If we want to play a target man who links up play, we will need to look beyond JDB - as that’s not his skill set – and beyond Yann who looks to be declining.