Stolen from Fiverweb done by a chap called TWar. Interesting read and obvs lots of effort.
A statistical analysis of the 20/21 Southampton season: What went right and then went horribly wrong in quick succession?
In this post I will seek to analyse our season, which players did well, which didn’t, what we could require in the summer (I intend to do a later post statistically analysing summer targets/signings if people enjoy this), and possible reasons for our end-of-season collapse. Most data will be acquired from OPTA stats via a subscription to fantasyfootballscout.com, when other data is presented it will be sourced. Not everyone wants to read a boring long post, so the main interesting points will be bolded and underlined.
Squad analysis
Goalkeepers
I considered the idea of putting strikers first and goalkeepers last in an attempt to open with positives but lets not beat around the bush. Both our starting goalkeepers have been, statistically, woeful .
The most important job of a goalkeeper is to make saves. McCarthy has the joint lowest goals prevented per 90 in the league (-0.32) , joint with Guiata of Crystal palace, of any keeper in the league that has played over 500 minutes. Ryan is lower in total as well and has technically played over 500 minutes but across two clubs which makes the number hard to judge. From this, we see that his shot stopping is dreadful. Forster is better with a goals prevented per 90 of (-0.14) which still puts him in the bottom 25% but is at least not an outlier in how poor he has been.
So the obvious answer is to play Forster? Lets look into that. Forster is betrayed by the type of save he makes. Forster has made 20 saves this season with a pretty dreadful save % of 64.5% (McCarthy is 63.4%). Of those saves however, not a single one was a catch, with 14 parries and 6 punches. For most goalkeepers catches make up about 25% of their saves, Forster has 0%. He constantly puts us in danger with his poor handling skills, pushing the ball out for set-pieces or dangerously back into play. Forsters recoveries per 90 is 7.05, is only better than Ederson, Mendy, Schmeichel, Patricio, and both manchester united goalkeepers. That may seem like prestigious company, however these goalkeepers have low recoveries due to their teams high possession and low amount of time with the ball in their half (with the exception of Patricio who also is very poor in this area). Other goal keepers who face a similar number of balls into the box as Forster are significantly ahead. McCarthy, for example, makes 7.48 recoveries per 90 which is still poor compared to most goalkeepers adjusting for possession, but better than Forster. From this we can see Forsters ability to claim the ball really lets him down, and justifies McCarthy having to play against cross-heavy teams.
Finally we look to distribution. Our side relies quite heavily on Vestergaard to make long balls from the back. In fact in the closing games of the season McCarthy was taking goal kicks to Vestergaard who was stood a couple of feet away from him, who then took the goal kick for him. This is not normal, often a short goal kick is taken but normally the goalkeeper is sufficiently trusted to pick a pass incase the CBs are marked, we do not trust McCarthy to do this, and hence he no longer takes his own goal kicks. For this we turn to FBRef for info as FFS didn’t have details. This resource only looks at primary goalkeepers for a team so we only have stats on McCarthy, however his launched ball (over 40 yards) completion % is a shameful 29.5%, rock bottom of the league and a full 16% lower than the median result. Whilst this website doesn’t provide details on Forster, it does display Southampton’s over all launch completion % which is 29.2%, this is EVEN WORSE, showing Forster is even worse at longer passing than McCarthy. We are still rock bottom of the league by this metric. Our need for a passer, like Vestergaard, is artificially larger than it should be as both our goalkeepers are embarrassingly bad at passing compared to the rest of the league.
Conclusion
Our goalkeepers are both dreadful, McCarthy is the worst at stopping shots in the league and Forster can’t claim a ball or keep hold at all, whilst also being a well below average shot stopper. Both are worse kickers than anyone else in the league and their divergence from the mean when it comes to distribution is, frankly, staggering.
Centerbacks
This section provides more positivity than the previous, however it also offers the biggest indicator as to the reasons behind our drop in form of any. I am separately talking about centerbacks and fullbacks this season owing to having a lot to discuss for both and them having very different roles.
A lot of people noticed the first sign of our decline was the loss of Vestergaard from this side. But was he that important, and was he really performing that much better than his replacement? Yes.
A commonly implemented strategy for us this season was to funnel attacking players out wide with narrow fullbacks and deep sitting defensive midfielders and then to win the ball in the air when the crosses come in and rely on the excellent 1v1 defending of KWP and Bertrand to ensure that players couldn’t run into the box from wide and play along the floor. A clever, if simple way of defending. This relies on three major facts to be true:
- The fullbacks must be good at 1v1 defending
- Our midfield must break up play on the edge of the box
- The centerbacks must be able to win the ball in the air.
We shall move onto the first two points in the next sections but first lets look at our centerbacks in the air. Vestergaard won 3.26 aerial duels per 90, putting him in 11th for CBs in the league (who played over 1000 minutes). Vestergaard does not carry the full burden of aerial defending however. Bednarek wins 2.64 aerial duels per 90, still putting him very respectably in the top 25 centerbacks in the league.
Here is, however, where the problems start. Stephens is dreadful in the air. While Vestergaard and Bednarek achieved 3.26 and 2.64 aerial wins per 90, Stephens made a much less impressive 1.12. This makes puts him very low in the league for this stat. His win % is 52%, with Vestergaard and Bednarek achieving 64.5% and 64.2% respectively. As Salisu was out, losing Vestergaard completely changed how we had to defend. Stephens very poor aerial ability ment we could no longer rely on frustrating teams by forcing them to launch the ball into the box, as now that is a viable strategy. So what about Salisu, does he have what it takes to step up if/when Vestergaard leaves. Salisu wins 2.13 aerial duels per 90, not on the same level as Vestergaard or Bednarek but significantly more than Stephens.
It’s not all about aerial duels though, sometimes the fullbacks or CMs won’t be able to screen threats to the CBs and they will need to act on the ground. Vestergaard has a tackles won % on the ground of 74%, this is significantly higher than Bednarek or Stephens 60% and 61% respectively . This means that if Vestergaard is forced to be the last ditch option then he is a lot more reliable at it than his replacement or Bednarek, yet another reason his loss really hurt us. Salisu, as those who have watched him have noticed, is excellent at tackling on the ground. He has a tackle win % of 82%, much higher than every other player in our team. If he can learn to be as effective in the air as the other two, and he’s not miles off, he will be a superstar.
Conclusion
We REALLY missed Vestergaard. Stephens is worse in almost every regard, especially at defending in the air which is so key to how we play. Without Vestergaard, and as said earlier, with keepers who cannot claim the ball well at all, we are not at all well served by funnelling players out wide as we did for the first half of the season. Bednarek may be poor at tackling and have a gaff in him but dropping him for Stephens would be very illadvised considering his dreadful ability at heading and not much better anything else. Salisu is a star in the making who just needs more pitch time.
Fullbacks
Our fullbacks, as stated in the previous section are essential to how we play. In this section we will discuss the decline of Ryan Bertrand and arguably our most important loss between the first and second halves of the season, Kyle Walker-Peters.
First, let’s discuss Ryan Bertrand. Bertrand has been one of our most reliable and important players over the last five years of the club. He has in the past been a rock solid stalwart at the back and a genuine threat going forward, but it is obvious he is in decline. Figure 1 shows a plot of Bertrands performance since arrival, xGI numbers for 2015/16 weren’t available and were hence extrapolated backwards (this is not an accurate method, do not pay too much head).
Figure 1 - Bertrands decline over time, CBI is clearances + blocks + interceptions
As you can see from this plot, Bertrand has steadily declined with regards to defensive output. He still puts up ok numbers but nowhere near the league leading quality he provided in 2015/16. Offensively we can really see what has changed with Bertrand this season specifically. He has dropped by a huge degree since last season, providing less than half the expected goal involvements of last season and well below his average since 17/18 which is the last source of accurate data. Bertrands decline in defence and attack are clear and pronounced and we are correct to move him on, with offense being the biggest decline this season specifically.
Kyle Walker-Peters is a great player, and a good shout for our most important player of the season. His tackles won % is second in the team after Salisu at 78%, this is amongst the highest in the league for players with over 1000 minutes. Going forward, however, he is a bit of a non-entity with an expected goal involvement of 0.07, lower than Bertrand. Owing to this, with Bertrands decline our fullbacks offer very little going forward. This is an issue we will no doubt be looking to address in the summer, and a possible reason why JWP was brought into RB when we were chasing the game against West Ham in the final game of the season. So how damaging was his absence? Well, as stated above the 1v1 defending ability of our fullbacks is key to how we defend and our ability to show players outside. As stated earlier, our two replacements in that role, Bednarek and Stephens, have a tackle % of ~60% each to KWP’s 78%, they were not up to the task of 1v1 defending which was required of our replacement RB.
I also tested a personal hypothesis of mine, that KWP was rushed back from injury too early and wasn’t fully fit. I believe this because after coming back he looked much less effective and had a couple of other spells out the side with flare ups. Before his injury KWP had a tackle % 77% and a CBI (see fig caption for definition) of 3.34 per 90. After his injury his tackle % stayed reasonably consistent at 79% but his CBI dropped to 2.9. Not conclusive evidence but potentially indicative of a niggling injury.
Conclusions
Bertrands decline defensively has been slow and steady, his decline in attack has been swift. He needed to be replaced. KWP offers little going forward, so without Bertrand carrying that burden we don’t create enough from fullback. He is however excellent defensively 1v1, something neither Bednarek or Stephens can say, therefore they were both woefully inadequate to replace him. We need quality at LB and depth in the summer. We already knew this, but the numbers support it.
Central Midfielders
JWP had an excellent season again. He lead the league for goals from set pieces and assists from set pieces, was second in the league for successful corners, and first for accuracy on target (over 10 taken). As a holding midfielder he also has done great with defensive output. He is top 14th for aerial duels won for midfielders in the premier league, 13th for tackles won, and 2nd for recoveries. I could go on about JWP but I don’t think it’s very productive to keep pointing out how good he is, and this review is more about where our season turned and went wrong. So we really should analyse his partners. Only two players won more tackles per 90 than Romeu this season in midfielder - Ndidi and Djenepo (surprisingly) (again only including people who have played over 1000 minutes, top in the league is Jankewitz). Romeu was also formidable for recoveries, getting 7th most recoveries in the league per 90.
So, after injury, was Diallo up to doing the job of Romeu? No. Below is a table of Diallo vs Romeu for defensive ability:
As can be seen, Romeu is well ahead on almost every metric. Diallo is a better passer than Romeu with a pass % of 85% (only bettered in our team by JWP with 86%) to Romeu’s 83%, and creates more chances than Romeu (0.73 vs 0.69 per 90) but he will not fill in his job when it comes to winning the ball back.
Conclusion
JWP is brilliant. Diallo can pass well be he is not doing the job of Romeu defensively and when Romeu is out we will really miss him.
Attacking midfield
Attacking midfield was, statistically speaking, our second biggest weak spot this season. For expected goal involvement per 90, we didn’t have a single player in the top 50 in the league for midfielders. Our best midfielder for creating chances was Ward-Prowse, who primarily operates as a holding midfielder, which is very poor. Below is a table of different attacking midfielders for our side and how they performed on different attacking attributes. In this CC stands for chance created and FTSP stands for final third pass completed, finishing is calculated by G - xG. This are all per 90.
One surprising stat here is the creative output of Redmond. Redmond tops our attacking players for expected assists, chances created, and final third successful passes . This could be due to his playing striker more often than some others however. Armstrong is our best finisher and our best assister on paper, and carries good goal threat. It is important to note for Armstrong that he played a significant number of games as a center mid and, therefore, his attacking stats are maybe slightly misleading. Walcott is also very creative, making good chances and getting a good number of assists. Minimino has more goal threat than our other players but is not very useful elsewhere and his assist potential is very low. Djenepo and Tella are not competing at the moment, they look good but statistically speaking and without confirmation bias it does make sense why Ralph benched Tella for Redmond consistently based on their respective outputs.
Conclusion
We need improvement in this area. No one is very good, Armstrong puts up OK numbers and finishes well which inflate them somewhat but we need to sign a good attacking mid. Minimino may be an ok option for scoring goals but that could just be sample size. Tella and Djenepo need more practice, neither is at a top level yet.
Strikers
Finally, that brings us on to strikers. Danny Ings is still very good. He put up 0.48 goals per 90 this season, joint 6th with Calvert-Lewin and Bamford for players who have played over 1000 minutes. He is the second best finisher in the league when it comes to strikers (behind Iheanacho) overperforming xG by 0.13 goals per 90 . If he had stayed fit all season he would have been very close to another 20 goal season. His injury likely cost us 7.6 goals, which would have been very meaningful in making our way up the table. We should try to keep Danny at all costs, when he is fit he is CL level atleast and has been for two seasons.
Adams has had an ok season too. He has scored 9 goals, which puts him 14th in the league. His finishing has again been pretty poor, underperforming xG by 0.85. He is 13th best for xG in the league, so it does appear his finishing isn’t hindering him as much as it seems however. Adams is a superb creative force, getting the 6th most expected assists per 90 of any forward in the league this stat shows he is contributing greatly in his own right and not just riding Ings’ coat tails
Our other forwards, Redmond and Tella, achieved 0.1 and 0.05 xA per 90. Adams also achieved 0.1 which shows Redmond is not as far behind Adams creatively as some might think , Adams is overperfoming his expected assists by 0.06 whereas Redmond is overperforming by 0.05 giving Adams a slightly better actual assist total. Ings also puts up 0.05 expected assists per 90, adding further credence to the idea he is the main man scoringwise and Adams/Redmond are very much in a supporting role.
Adams scores 0.29 goals per game from an expected 0.33, which is far in excess of Redmonds 0.1 goals from 0.15 expected. This shows that while Redmond rivals Che for assist potential, with scoring he has some ways to go. Tella is not near either on goals or assists as can be seen in the above table.
Conclusion
Ings is great and we must keep him and keep him fit, could easily tear it up at a top side. Adams is surprisingly proficient at assisting and is putting up fine numbers goalwise for a striker in midtable, to push on we need someone next to him atleast equally as good. Redmond assists well but doesn’t score, Tella has a way to go.
Overall Conclusion and attempted answer to “where did it all go wrong”
One narrative persists all the way through this analysis: our second string of players are bad. We have some obvious weaknesses in our first team this season (goalkeeper, attacking mid, leftback) but the biggest thing from all this is just how far behind our back ups are. Stephens is miles behind Vestergaard in basically every regard, Bednarek is a laughable replacement for the role KWP does, Diallo is not even close to equipped to do Romeu’s job, our attacking mids all have massive gaps, and Redmond has no where near the goal output to replace Ings.
I suggest from this that in the first half of the season we were a very delicate machine that worked only when every cog was in place. When the cogs broke we replaced them with elastic bands and chewing gum and by the time the key players were back, we had nothing to play for, the players did not look at all fit as if they had been rushed back, and our confidence was in a toilet.
I hope you enjoyed reading this, or skimming it as most probably will as it is a massive essay. The facts I posted here are true (to the best of my knowledge and ability), but my opinions and editorialising are not and you are welcome and encouraged to draw your own conclusions from what I have presented