Stats Corner: Expected Goals

Stats Corner: Expected Goals

In a foot­ball world stuffed with tech­nol­o­gy and big data, it can be hard to untan­gle from the impres­sive amount of stats, index­es, and algo­rithms that nowa­days rep­re­sent the basis for the study and analy­sis of games, train­ing, and play­ers. In fact, in the last years – also thanks to tools like Wyscout – man­agers, play­ers, coach­ing staffs, ref­er­ees, tal­ent scouts, agents, jour­nal­ists and many oth­er pro­fes­sion­als of the Beautiful Game inte­grat­ed Advanced Metrics and Big Data in their dai­ly work, start­ing what we could define Football 2.0.

The process didn’t hap­pen overnight, and there still is some con­fu­sion about which met­rics and index­es are the most rel­e­vant to the analy­sis of the game and what they actu­al­ly rep­re­sent. Just con­sid­er­ing the Wyscout plat­form, there are more than 150 stats, both for clubs and sin­gle play­ers, and many of them can be con­sid­ered as Advanced Metrics, that are deep­er stats that ana­lyze every part of the game in detail.

To make every­thing clear­er, we decid­ed to inau­gu­rate the Stats Corner” col­umn here on Wyscout Blog, pre­sent­ing some of our Advanced Metrics on a reg­u­lar basis, explain­ing how they are cal­cu­lat­ed and what they mean. We decid­ed to start with the Expected Goals – also known as xG – a sta­tis­ti­cal index that eval­u­ate the qual­i­ty of a play­er or a team’s goal-scor­ing oppor­tu­ni­ties. Here’s the def­i­n­i­tion from the Wyscout platform:

Expected Goals is a met­ric that assigns to every shot a prob­a­bil­i­ty to be con­vert­ed into a goal. This prob­a­bil­i­ty is cal­cu­lat­ed from the cor­re­la­tion between the chance to score accord­ing to spe­cif­ic char­ac­ter­is­tics of the shot – the posi­tion, the type of the shot, the game sit­u­a­tion, the type of assist, the dis­tance to the goal – and the respec­tive stats archive on Wyscout.

Pretty clear, right? In case you need fur­ther expla­na­tion, here you have. First of all, Expected Goals is an index; that means is not a sin­gle stat but a group of stats that are cor­re­lat­ed and eval­u­at­ed with­in a spe­cif­ic algo­rithm. So, in this case, the Expected” word can be a false friend. In fact, xG is not a pro­vi­sion­al data but an index based on a con­clud­ed event. Here are a cou­ple of prac­ti­cal exam­ples to make every­thing clearer.

Low Expected Goals Index

Let’s start with a sit­u­a­tion with a very low xG index. That means, sta­tis­ti­cal­ly speak­ing, a shot that has been con­vert­ed into a goal that – at least accord­ing to our stats archive – shouldn’t have had a sin­gle pos­si­bil­i­ty to become so.

Let’s take a look at Manchester City’s sec­ond goal in the recent clash against Maurizio Sarri’s Chelsea at the Etihad Stadium.

At the 13th minute of the first half, Citizens’ strik­er Sergio Agüero shots towards the goal from far beyond the box, and also from an edged posi­tion. Even if shoot­ing with his strong foot – the right one – sta­tis­ti­cal­ly Agüero has very few chances to score that goal. How few? According to Wyscout’s algo­rithm, Kun’s goal had a 0.03 xG val­ue. Almost impossible.

As we were say­ing, this val­ue is cal­cu­lat­ed by con­sid­er­ing the char­ac­ter­is­tics of the shot – posi­tion, dis­tance to the goal, the play’s build-up, the shoot­ing foot, etc – and cor­re­lat­ing them with the stats archive about sim­i­lar shots. Basically, ground­ing on exist­ing sta­tis­ti­cal data, Agüero’s shot had the 3% of chances to become a goal.

High Expected Goals index

Switching to an oppo­site exam­ple, let’s take a look at a goal with a high­er chance of con­ver­sion. It’s November 2018 and, at the Estadio Ramón Sánchez Pizjuán, Sevilla hosts Liverpool in the UEFA Champions League group stage. At the 30th minute, with the Reds already lead­ing 1 – 0, this happens.

Following a lat­er­al dri­ve by Sadio Mané, which con­clu­sion is blocked by Sevilla’s goal­keep­er, Roberto Bobby” Firmino finds him­self alone in front of an uncov­ered goal, eas­i­ly scor­ing after con­trol­ling the ball and also indulging him­self with a no-look fin­ish­ing. But the Brazilian’s con­fi­dence was more than jus­ti­fied. In fact, the xG index for that shot was 0.85. Almost impos­si­ble… to miss (always in cor­re­la­tion with the stats archive about sim­i­lar shots).

Of course, these are just two extreme shades among the hun­dreds of vari­ables that data like Expected Goals com­pre­hends. Just think about the (almost) infi­nite shot pos­si­bil­i­ties that can hap­pen dur­ing a foot­ball game – from penal­ties to impos­si­ble bicy­cle kicks, from lucky shots to unthink­able trajectories.

So, what are Expected Goals for? Ultimately, is a met­ric cal­cu­lat­ed in order to ana­lyze the result of the game with more objec­tiv­i­ty, remov­ing – as much as pos­si­ble – its most casu­al part. If you fol­low Wyscout offi­cial Instagram pro­file, for exam­ple, you may have seen our Expected Results of the Weekend” col­umn, where we take some results from the pre­vi­ous foot­ball week­end and we com­pare them with the Expected Results”, that would be the sta­tis­ti­cal results accord­ing to our xG index.

If you take the 3 – 3 draw between Atalanta and AS Roma – 21st Serie A game­day – it can acquire a very dif­fer­ent mean­ing if com­pared to the Expected Goal val­ue: 2.57 for Atalanta, 1.55 for Roma. Other than the rough-and-ready analy­sis – Atalanta deserved the win” – xG gives the result a more pre­cise val­ue, both in terms of goal scor­ing oppor­tu­ni­ties cre­at­ed, with the Nerazzurri way more active in front of the oppo­nents’ goal, and in terms of con­crete­ness, with the Giallorossi capa­ble of scor­ing the same goals with few­er occa­sions to do so.

Then no, the Expected Goal index won’t tell you which team real­ly deserved the win, as that’s always the team that actu­al­ly won. But it will help you – whether you are a coach, a match ana­lyst, a scout or any oth­er pro­fes­sion­al in foot­ball – to bet­ter ana­lyze the goal scor­ing oppor­tu­ni­ties gen­er­at­ed by a play­er or a team dur­ing a match. That’s rather good, isn’t it?