You’ve probably heard of xG models.

The xG stands for expected goals, and these models divide online armchair statisticians almost as much as Arsene Wenger splits the Arsenal fanbase. A topic of much contention then.

The principle is fairly simple: calculating the chance of a shot being scored. The details are somewhat more complicated though. They typically look at factors like where it was taken from, what kind of pass led to the chance, and what part of the body took the shot. Once this is all complete, it allows us to argue incessantly about how many goals a team “should” have scored, and therefore who deserved to win a game, irrespective of result.

Or it allows us to criticise a particular striker’s finishing. (Or accept that they actually have one of the best conversion rates in Europe.)

It even allows us to understand whether frequent comments about Arsenal trying to “walk the ball into the net” actually have any merit. It might seem like it sometimes, but there’s a difference between aiming for volume of chances versus quality.


Importantly, though, expected goals allow us to look at a more advanced metric than shot conversion. Just as in the real world, the xG model attempts to consider that not all chances are equal. After all, that’s a fairly important qualification when looking at performance.

Anyone remember the year Andy Johnson was the Premier League’s second top scorer with 21 goals? 11 of them were penalties. Compare and contrast with the man who topped the list, a certain Thierry Henry. 25 goals. 0 penalties. Hardly comparable.

These models aren’t perfect, as they don’t take into account the quality of the shot. For example, Kevin de Bruyne so ably demonstrated last weekend that not everyone can score a simple tap-in. Certainly not as composedly as Mesut Ozil did against West Ham…

However, they’re still a step up from a pure conversion metric.

So yes, xG has a value. I’m a bit of a geek – I have a physics degree and my day job involves working with numbers – so I love these more advanced metrics.

Turning the tables on goalkeepers

Historically these numbers have all looked at the success or otherwise of teams from an attacking point of view. Imagine my excitement then, when I saw Sky Sports publish “xC” (or expected goals conceded) statistics this week.

A chance to take a more refined look into whether the perceived best goalkeepers really are the best goalkeepers.

The data, provided by Opta, considers the chance of the shot going in, similar to the xG models above, but the numbers look at goalkeepers instead of the attacking side. Based on the quality of chances and shots against them, how many goals should they have conceded?

Anyone who has watched Burnley this season, will nod knowingly when they see that Tom Heaton is top of the pile. The clarets stopper conceded 19 goals against a modelled number of 24.06. In real terms, that means that against the law of averages, Heaton has so far saved Burnley five goals, which strikers would usually have scored.

In the top six Premier League teams, Hugo Lloris has saved Spurs a goal on their modelled goals against, and Loris Karius is just about breaking even, having conceded 10 goals against an xC of 10.26.

However, David de Gea is in negative numbers, having conceded 16 goals against an expectation of 14.92, Claudio Bravo has shipped 12 against a projection of 10.75, and Thibaut Courtois has practically been waving the ball through 11 times against the modelled 8.93.

You feel for David Marshall of Hull City. He’s either faced a number of low percentage wonder goals which are unsaveable, or he’s having a nightmare, based on 22 goals conceded against 14.49 expected.

Arsenal’s main man

And what of Arsenal’s ‘keeper? Petr Cech is among those with positive numbers, conceding 14 goals against an expected 16.41, worth 2.41 goals to us this season. That places him third overall.

I found this particularly interesting, as the eye test suggests to me that Cech isn’t having his best season. Perhaps that is where the statistics help to add another lens.

Many of us keyboard warriors have written on his near post deficiencies, but I’ve also felt that our number 33 hasn’t been keeping out as many freekicks as I might perhaps have liked. However, it’s easy to become blinded by the perceived errors. They might even make us overlook the excellent saves from open play in particular where the ball goes across him.

It perhaps goes to show that no ‘keeper is perfect, and while he might have a near post weakness, he might also make up for it by saving more than average at his far post.

This is where it might be particularly interesting to overlay where the shot was aimed into the model. Perhaps in (many) years to come we’ll have access to that kind of insight!

Arsenal’s “other man”

As always, it’s worth remembering what the statistics don’t tell us.

Obviously David Ospina didn’t feature in the publication, since he hasn’t started a single Premier League game this year. That’s a gap which makes it somewhat difficult to compare our two keepers on this basis.

I have no idea what his “xC” numbers would look like based on the European games he has played. The sample size is pretty small regardless. I’m hardly a fan, as regular readers will know in any case, but whether the numbers look positive or negative, there’s still a hole in the story nonetheless.

Expected goals conceded doesn’t help us to understand mistakes which lead to chances. Nor does it consider aerial blunders like catching a cross and then throwing it in your own net. Or understand the impact of under-strength kicking inviting pressure onto the team. Or the panic that can afflict the back four without Cech’s calming influence.

For those, we rely instead on the eye test – what we see with our own two eyes – for better or worse. In my case, my optician tells me my -7 vision is very much “for worse”!

All I can hope is that we don’t have to use that eye test in the Premier League this season!