# Pierre-Emerick Aubameyang doesn't need your expected goals

**By Dan Altman, creator of smarterscout.com**

Football analytics were very different back in 2014. Stats that we don't hear much about today were popular, and most of them looked at actions in isolation rather than underpinning a model of the game. Writing for the now defunct Bloomberg Sports website in February of that year, I offered a foundation for such a model by breaking down goal difference into component parts. Even today, people tend to ignore a couple of those parts – but not the people who follow Pierre-Emerick Aubameyang.

The idea behind my old article (preserved here on the North Yard Analytics website) was to divide up goal difference into parts that were measurable and either random or persistent. Here's the final equation for a team's goal difference over a match or any other period:

*GD = OGD + xGD + (N *· *eSh) + (M *· *eSt)*

*GD* is goal difference, and *OGD* is own goal difference – or, if you prefer, own goal plus penalty goal difference. *xGD* is expected goal difference, based on estimates of the chances of scoring for every shot the team took and faced. *N* is the number of shots the team took during the period, and *eSh* is an error term conveying any additional success in finishing. *M* is the number of shots the team conceded, and *eSt* is an error term signifying unexpected success in shotstopping.

My feeling at the time was that *OGD* was fairly random, while *xGD* might be persistent. I called *eSh* and *eSt* error terms, because back then I hadn't proven to myself that finishing skill and shotstopping skill were persistent. Now I know that they can be; that's why we rate them on smarterscout.com. They're also really important, which brings us to Aubameyang:

If we only looked at the expected goals from Aubameyang's shots – in other words, our estimates of the chances that a generic Premier League striker would score from his chances – then we'd see a striker in decline. In 2018-19, the our estimate of the average chance of scoring from the shots he took at CF was 19%. In 2019-20, that estimate fell to 15%.

But even though Aubameyang's average shot quality fell, his conversion rate rose, from 16% to 22%. A similar thing happened for him at LW. Our estimate of his average shot quality fell there, too, from 16% to 11%. Yet his conversion rate fell by a smaller amount, from 29% to 25%.

That's why Aubameyang actually scored at a higher rate in the most recent season. He had 16 non-penalty goals across 2,261' at the two positions in 2019-20, compared with 15 non-penalty goals across 2,435' in 2018-19. Aubameyang was taking worse shots – indeed, his overall attacking output in our shot creation model was falling – yet he was becoming a better finisher. One of the components of the equation above that's not based on expected goals was just as important as the part that was.

This conclusion is borne out in our metric for finishing skill of non-headers in open play. As you can read in our FAQ, the metric rates strikers and goalkeepers simultaneously so that every shot is benchmarked not just for a generic striker's average chance of scoring, but also for the relative skill levels of the striker and goalkeeper involved in the play.

The raw metric looks at a striker's chance of scoring from a basket of shots where a generic player would have a 50% chance of hitting the back of the net at least once. When a player switches leagues, we publish an adjusted rating from his old league until we have enough data from his new league, where he starts at par. Here's how Aubameyang developed since he came to the Premier League:

Aubameyang started well enough, but his finishing began to decline midway through the 2018-19 season and hit bottom towards the end of that campaign. Then it turned around and kept climbing – not even stopping for the pandemic-induced break – until now, when it may finally be starting to plateau. At this point, Aubameyang would be expected to score 58% of the time from a basket of shots where a generic player would have a 50% chance. And that's just one type of shot.

What changed for Aubameyang? We can start to answer this question using data. First, did the locus of his shots just change to an area that was more amenable to his finishing? Let's look his shot map for non-headers in open play at CF/ST in 2018-19:

Aubameyang had quite a few misses in the area roughly level with the penalty spot. Now look what happened in 2019-20 when he played the same position:

Those misses pretty much disappeared in 2019-20. To be sure, Aubameyang played more minutes at CF/ST in the earlier season. But his frequency of shooting did still fall in 2019-20, with his style rating for shooting falling from 96 to 77:

So it looks like Aubameyang became more selective in his shooting. At LW, by contrast, his shooting rate went up slightly – but mainly just inside the box after coming through the left channel, which was also the site of his trademark runs when he was at Borussia Dortmund.

Why did this happen? Did it help that Aubameyang had the summer off in 2019, after Gabon failed to qualify for the Africa Cup of Nations? Did Unai Emery and his staff coach him to be more patient? Or was it because Aubameyang was entering the last year of his contract with Arsenal, so he spent extra effort preparing to put himself in the shop window?

I can't tell you the definitive answer with these numbers and shot maps. But the data have already helped us a lot by getting us to ask the right questions. They're the sorts of questions I'd be asking if I were thinking about signing Aubameyang this summer. And I wouldn't be asking them just because of his expected goals, though expected goals were baked into the rating for finishing skill that got me here. To see the whole picture, I had to see the whole equation, too.

**Would you like to have your player featured in the playerfocus on our homepage, for thousands of users and football professionals to see?** Just click the Contact link at the bottom of this page to email us your request.

[Photo: Chensiyuan]