Dyche, Expected Goals and The Season Start

There has been a lively, and interesting debate that has begun over Expected Goals, in light of Everton’s performances thus far this season. In short, there is a big gap between Everton’s expected points, and the points achieved, which has sparked some worthwhile discussion which should be addressed. In part to put some context in terms of what the numbers mean, what they offer and what they ultimately don’t.

As for a full disclosure, I am a big fan of Expected Goals data. It is objective, it is consistent and it came present the picture of a game devoid of emotion, context and any human bias that we all suffer from. As human’s, it is in our nature to look to create a narrative around a match, which is largely influenced by the result of the match. Performances that end in wins, are more likely to be crafted as good performances, whereas those that result in defeats can be cast as a poor display. It’s not a criticism, it’s an inevitable part of being a football fan and one we all partake in. Xg offers a bit of a challenge to that, as it skips those biases. It makes an objective judgement, primarily on the shots you take and the position you take them from.

It is perhaps better to state what Expected goals doesn’t do. It doesn’t tell you how well you play. You can have a higher Expected goals playing badly, just as you can win a game without always playing well. It is also the case that the process of going from an expected goal, to a goal is not without necessary skill, indeed it is likely the most important skill within football. The ability to finish chances at one end, and to prevent goals at the other. The ability of your goalkeeper, and strikers is not taken into account, that these are arguably the most important positions on the pitch. It also doesn’t fully take into account game management by astute managers. There is a huge amount of context to add when evaluating Xg.

That being said, even with those criticisms, it offers a valuable resource. Typically, sides tend to revert very closely to their Xg over longer periods. Last season, the biggest adjustment from Understat to the two numbers was Wolves, who finished 19th on Expected points, but 13th in the PL. The biggest gap was Leicester, who finished 11 points lower than they ought to finish. It is a worthwhile underlying measure at showing effectiveness in chance creation, and giving away chances.

While Everton’s performance of 7 points from 8 games is disappointing, the underlining metrics are very strong. They are 5th in terms of Xg created, but have the lowest chance conversion, but have one of the lowest conversion rates in the league. The return of Calvert Lewin and the signing of Beto has helped enormously in resolving this pattern, and suggests that this gap start to be closed over the coming weeks, if the XG performance stands up.

The XG data has been collected since 2014. In terms of XG performance, this season is the strongest performance after 8 games. Using rounding up/down in terms of goals scored, Everton have deserved to win 6 games, and lose 2 (with 1 of the losses quite a close game) showing 18 points if Xg was reflected in terms of standard finishing. On 2 occasions, (Koeman’s first season and Benitez’s season) we had 15 Xpoints from the first 8 games. On both occasions, performance levels really dropped immediately following those 8 games, which is very much Dyche’s challenge as the fixtures turn. While you can quibble the specifics of the XG/XP created, the general trend is that the underlining metrics are strong. You sense either they drop, or the points accumulation grows.

It is partially why I have been sympathetic to Dyche. The creation of opportunities has been good. In some ways this is the strength of Xg, while the general consensus is Dyche is a defensive manager, without much attacking plan or prowess, the reality is that Everton create lots of chances. The issue is how clinical we have been, and the qualitative factor of having a proper number 9 (Beto) and a return to fitness of a very good number 9 (Calvert Lewin) suggests it is both easier to create chances, and they can hopefully start to convert more. However, whatever criticisms of Dyche’s methods exist, he has proven surprisingly effective at making chances. It would be a huge risk to adopt a radically different method of playing.

The reality seems to be that the power and improved intelligence of Calvert Lewin, alongside the mobility and power of Doucoure has proven very difficult to handle for opposing teams. Dyche has done the simple thing, which is to maximise their strengths, play more direct and ultimately get the ball forward to them as quickly as possible.

If there is a criticism of Dyche, beyond ensuring the side finish chances better, it would be the defence. When he came in, I had assumed he would be an Allardyce+ type manager. Low XG, low XG against, quite effective conversion of performances into points. The reality is, we have given quite a bit of goals against, sitting 11th in the table, against what is quite a favourable set of teams. This needs to improve as the season goes on and should be a focus for the manager.

However, the manager deserves time to see if he can turn around some promising underlying numbers into positive results.

Alongside some good team metrics, there are also some promising individual Xg data, albeit with the caveat that it is a smaller sample set. In terms of Xg, I was always struck by how much DCL dropped off under Lampard. He was at 0.27 Xg per 90 minutes under him. When Dyche came in, this jumped to 0.61 per 90. Essentially, he became far more effective under Dyche, who seemed to know how to utilise him better. So far this season this has jumped to 0.85 Xg per 90 minutes. This is strikingly good. Of those who have played 250+ (as DCL has) only Haaland &(1.09) Nunez/Isak, (1.11) have higher. In Nunez case these impressive numbers are slightly bumped up by having lots of sub appearances, which tends to inflate numbers.

DCL’s are very good. For context, in the 2 season’s Ancelotti was at the club (prior to injury problems) he played 2871 & 2621 minutes. If they were to be repeated, he would score 24 an 22 goals respectively. That remains a huge challenge, and he may begin to drop down on Xg with tougher fixtures coming, but it gives a glimpse at numbers centre forwards haven’t been able to get anywhere near in his absence. Even if he only got half of that output (11/12 goals) the impact would be huge, and enough to lead Everton well away from the relegation drop zone. I did state at the start of the previous season, if Calvert Lewin is fit Everton will stay up comfortably, if he isn’t it will be very dicey, and I have the same feeling for this season.

The numbers are striking compared to all other strikers for Everton in that time. Lukaku in his final 2 seasons got between 0.45 and 0.54 goals per 90 (the higher figure being Martinez’s last season) and Calvert Lewin himself was 0.57 in Carlo’s 2nd season. So, his 0.85 figure is very promising when placing it in the wider context of former Everton strikers. Of course, the brilliance of Lukaku was not just he would convert to that number, but that he would regularly play over 3000 minutes a season so was very resilient.

The other big plus thus far this season has been Doucoure, who himself is averaging 0.47 Xg per 90 this season against 0.19 last season. His numbers now have a slightly larger data set (678 minutes) and indicate a very effective 2nd striker. Again 0.47 per 90 is comparable to Lukaku’s output. In terms of the PL he is right up there with other midfielders, Saka has 0.58, Mbuemo 0.57, Bowen 0.46 Odegaard 0.33 & Soucek 0.41 being the notable comparison’s for him. There is again a question of fitness, though prior to joining Everton he had 3 seasons where he played over 3000 minutes (3324, 3061 & 3156) while his 2 at Everton (prior to Lampard) saw him play 2443 & 2556 minutes. Even if he could replicate those minutes he is suggested to score 11/12 goals, if he could do Watford numbers it would be 14/15 goals. Again the weaker opposition has to be factored in with Doucoure, but the hope would be that he could get towards double figures.

For Dyche, it could only be seen as 2 big wins for him. Doucoure was bizarrely ostracized by Lampard and Calvert Lewin was badly used and struggling with injuries. With Doucoure he not only made a big call to bring him back, he has made him a key fulcrum of the team in the way Tim Cahill was some 15 years previously, and he has put enormous faith and patience into Calvert Lewin. The style of play suits them both, and however tentative, both are showing initial promising signs.

You also have to spare a thought for Benitez, who himself looked to do something similar to Dyche, by playing Doucoure in a more advanced role (for Benitez more of an advanced midfielder to Dyche’s supporting striker) and make Calvert Lewin pivotal. As noted above, it started well for Benitez, but injuries to both greatly hampered what was likely quite a sound plan for Benitez. He didn’t have the flexibility, or ultimately the players to adopt a plan B but there was undoubtedly some misfortune for him.

The glib truth is Everton simply couldn’t afford two players of that caliber who were in their prime, with unblemished injury records. So there is no small amount of hoping that goes on with the injury record for both. However, if both stay fit, it’s hard to see Everton having any concerns with relegation, given the impact they have shown.

What is also pleasing, is other players also seem to be contributing re Xg. Danjuma is at 0.45 n Beto is at 0.44 so both also look a threat, albeit off smaller data sets than can be generalisable. But Everton pose more of a threat than in previous seasons, and look well covered to create chances.

This is also reflected in the Xa (expected assists) data. Since records were kept, we have a couple of seasons where Ross Barkley gets 0.22/0.24 Xa, Sigurdsson and Iwobi tend to be 0.15-0.2 range while James is at 0.27 per 90. This season has seen James Garner at 0.31 Xa and McNeil at 0.57 Xa. Again, it is a small data sample, against easier teams but if both players halved their output, it would still be very impressive. Garner who offers a lot else would be close to Sigurdsson/Iwobi levels, while McNeil would be the best creator we have have in the last 9 seasons. Given both(though specifically McNeil) offer an enormous amount of defensive output, which say a Barkley or a James didn’t, those numbers are striking and cause for cautious optimism.

So in conclusion, what do the underlying numbers tell us of the season so far? Firstly that Dyche has been a little unlucky, and probably deserves a bit of patience. The team is attacking effectively, but needs to get better at finishing. We remain very dependent upon Calvert Lewin and now Doucoure to stay fit. In Dyche’s new system, both players have made themselves key figures in an unusual turnaround. James Garner and McNeil show promise as players who can create for others. Finally, the defense needs to grow more consistent and give away less opportunities, particularly as the fixtures become more difficult.