Deconstructing Narratives: Arsenal shoot poorly in semifinals

It is that time again

Thursday, April 29, 2021.

This was the last time that Arsenal have scored a goal in a semi-final, it came in a 2-1 loss away against Villareal.

Since then Arsenal have played six matches and scored ZERO goals. Over these 6 matches they have produced 8.3 xG from my xG model.

They have played a 0-0 draw against Villarreal at home:

Drew 0-0 against Liverpool (playing most of the match with 10 men).

Lost 0-2 away against Liverpool

Lost 0-2 to Newcastle

Lost 2-0 away to Newcastle

Lost 0-1 to PSG

These are matches that span 4 seasons and feature a number of players that have changed over that time.

But, just how improbable is this?

Straight into it, this is an incredibly improbable outcome!

This is really pushing the bounds of things, over 3 absolute deviations from the expected mean.

This should happen in a tiny fraction of occurrences if xG is a good measure of the true estimate of chance quality. xG is not a perfect estimate for the quality of each chance, but we would not expect it to be off in such a systemic way here to be the main reason for this miss.

Arsenal have actually been a team that has over performed xG over this span by 40 goals compared to expected goals, in European Competitions plus Premier League.

This is also not exactly a situation where Arsenal have been reliant on a high volume of low quality shots to pump up the numbers here.

This is a sample of 68 shots, it has an average xG of 0.122 xG per shot (this is a bit higher than a typical Premier League team at 0.11 per shot). There are 9 shots here with an xG of 0.3 or higher, 14 shots with 0.2 or higher rating and just 30 of 68 that were rated under 0.05 xG.

So what’s up then?

But this isn’t purely luck, it is also something where it might be our pattern-seeking brains searching out something to tell a story. One possibility: Are Arsenal just turning into feeble shooters under the bright spotlight of a semifinal?

Well, in a way - kind of, yes.

Of the 68 shots here, just 14 were on target. That is a measly 20.6% on target rate and that is well below Arsenal’s average of 33% during this four-season stretch.

The estimate for the quality of these shots on target is also a pretty big drop off here at just 3.92 goals.

Simulating these shots gives the following outcomes:

So what the data measuring the quality of chances tells us is that average shooters should’ve been able to get around 8.3 goals in these matches. And the data also tells us that Arsenal have taken below average shots in this time, but still good enough to score either three or four goals in these matches (48.4% of outcomes), not none at all.

That’s the answer to the question — Arsenal have been shooting worse in semi-finals, but not so badly they should have zero goals. Article over, right?

WRONG!

Adam was thinking a little more deeply about the specific chances behind the data, and things started to crystallize a bit more.

First up: this 8.3 xG includes Gabriel Martinelli’s missed chance in the 40th minute, a toe poke at the back post that was off target but was also clearly offside. But, since Martinelli missed, the chance was never *officially* ruled out as offside, and wasn’t reviewed, because why would you? That means basically all models will still count it.

There is, of course, little sense in counting this chance. It ultimately affected proceedings no more than Mikel Merino’s headed goal, which turned a lower-xG opportunity into a high-xGOT outcome via a well-taken header but was chalked off. That should be considered shooting well in a semifinal, but won’t be anywhere in the numbers. So it is fair to remove this from our sample, or it least place a pretty heavy asterisk next to it.

Martinelli’s chance counts for around 0.8 xG in most of the models we’ve seen, but the Cannon Stats xG model puts it closer to 0.5. For the sake of conservatism, let’s call this a 0.5 xG chance, with 0.00 xGOT given he missed the target. Just like that, we’ve accounted for 0.5 of the 4.38-xG gap between Arsenal’s xG and xGOT, leaving a 3.88 xG underperformance in their shooting.

The other huge factor here: Arsenal shot really, really poorly in the first leg of the Carabao Cup semifinal against Newcastle. In that match, 23 shots were taken, and 3 ended up on target. On 2.99 Cannon Stats xG, Arsenal generated just 0.7 xGOT, an embarrassing 2.29-point gap.

Does this mean that Arsenal’s performance in the first leg should be written off and removed from the data? No, absolutely not. But it should be noted as a heavy influencer on the data, and factored into analysis that’s attempted to identify trends.

With that and the Martinelli play off the books, you’d be looking at 4.81 xG and 3.22 xGOT in the remaining five matches, a gap of 1.59 points between xG and xGOT. Those are numbers such that you’d say Arsenal haven’t shot well in semi-finals going back to 2021.

And that’s fair enough. But (LOL, here comes another twist) it’s probably worth pointing out the Villarreal tie’s role in all of this. With 0.94 xG and just 0.03 xGOT (what the hell?!) coming from that match, about 60% of the remaining gap is four years old. We’ve isolated by match in the below, for convenience:

It’s not worth the mental gymnastics here of removing Villarreal from the numbers, because that’s another match where Arsenal shot really poorly, and it did happen during the Mikel Arteta era.

But is it relevant? That’s up to the reader. That lineup featured our current Arsenal players — Saka, Ødegaard, Martinelli, Partey — and they combined to take three of Arsenal’s 14 shots, totaling .09 xG and 0.00 xGOT. So, yeah.

Either way, you’d say that based on the numbers, Arsenal have shot better than average in one of these six matches (Liverpool home), about average in three (Liverpool away, Newcastle away, PSG home) and very, very poorly in two (guess which).

The xGOT suggests that Arsenal should’ve had four goals in these matches, based on shot quality, location and typical keeper play. Not zero. So there is some element of bad luck, great keeper play, or both happening here, with bad shooting interspersed in select matches.

“But why always Arsenal?”

The reality is that hot keepers or lucky/unlucky breaks with regard to shooting end seasons, title pushes and Champions League runs all. the. time.

Arsenal were just the beneficiaries of this not one round ago against Real Madrid, a tie in which they won the xG battle about 3.7 to 1.9 but the real-life battle by a 5 to 1 margin. Why? One big reason was Arsenal shot well for the tie, grabbing 4.74 xGOT. Real Madrid was below average, with just 1.62 xGOT for the tie.

Elsewhere in the same knockouts, Bayern and Inter played a two-legged tie that ended 3.7 to 2.27 in Bayern’s favor on xG. Despite that, Inter advanced by a single goal, and a lot of that had to do with shooting, keeper play and friggin’ luck, for lack of a better word.

Inter’s 2.51 xGOT ended up accounting for four goals, while Bayern’s 3.84 xGOT (just slightly above-average quality in terms of shot execution) only resulted in three goals. Why? A couple of moments made the difference. One was Yann Sommer saving Harry Kane’s good header early in the first leg, a chance in which Kane nearly tripled the odds of scoring with the shot he took. In the second leg, Thomas Müller turned a 0.05 xG chance from the edge of the box into a 0.35 xGOT chance, but again Sommer saved it. Müller took another great shot in fifth minute of stoppage time, a header, that was rated 0.62 xGOT at just 0.22 xG. Sommer saved that, too.

On the other side, Inter capitalized on a couple of moments where young keeper Jonas Urbig probably could have done better. One was the first goal of the tie, a roofed shot from Lautaro Martinez that was rated 0.15 xG but 0.28 xG, still a saveable shot, but Urbig couldn’t stop it. Benjamin Pavard’s headed goal in the second leg was rated very similarly at 0.21 xG and 0.33 xGOT. As the numbers would otherwise indicate, outside of those two chances Inter actually shot *below average* for the tie.

If that doesn’t hit close enough to home, consider PSG-Aston Villa. That was a close tie over two legs, going 5-4 to the Parisians. And it was close on xG as well, 3.64-2.79 going the same direction.

But the quality of shots would actually have favored Aston Villa to advance. Their xGOT was 4.23 for the tie, compared to PSG’s 4.07. They shot especially dangerously in the second leg at home.

So, what the frick happened? Gianluigi Donnarumma.

The large Italian saved chances in the second tie worth the following xG/xGOT:

• 0.09/0.54 (Marcus Rashford 57’)

• 0.13/0.57 (Youri Tielemans 59’)

• 0.51/0.43 (Marco Asensio 70’)

Those are three chances that, based on how well they were taken, would’ve typically resulted in a goal and a half. Two of them were taken really, really well, statistically speaking. No goals, good keeper, season over.

Bad luck? Wrong place, wrong time? Variance? Call it whatever you want, but it remains incredibly frustrating when you’re on the receiving end. But in such a low-scoring sport, that’s just how it goes.