xG and Stats Mega Thread

[aaronaldo]

I thought I'd start a thread for us to discuss xG, xGOT, xA, etc trends throughout the season. I don't want to derail the more generic stats thread with this info. To start here is the explanation of the different terms and their meaning. I think xGOT is really interesting.

Expected Goals (xG)
Expected goals (or xG for short) is the likelihood that a shot will result in a goal. In other words, a measurement of how good the chance was.

A player or team's total xG in a match can give an indication of how many goals they could have expected to score, given the quality and quantity of the chances they created.

For each shot, an xG value from 0 to 1 is assigned, with a value of 1 representing a guaranteed goal. So, if a shot has an xG value of 0.4, that means you would expect a goal to be scored from that shot 4 out of 10 times. This is based on the shot angle, distance from goal, type of pass, header or shot, among other factors.

So, if a team loses 1-0, but they had an xG of 2.1, you can say they were unlucky to lose, as they could have been expected to score at least 2 goals from the chances they created.(Opta)

Expected Goals On Target (xGOT)
xGOT not only takes into account the quality of the chance, it also measures the quality of the shot. A value is assigned to the goalmouth location where the ball ended up, giving more credit to shots that end up in the corners compared with shots that go straight down the middle of the goal. This model is only for on-target shots given that if you don’t get your shot on target, there's a 0% chance that it will result in a goal. Essentially, xG is a pre-shot model while xGOT is a post-shot model. (Opta)

Expected Assists (xA)
Expected assists (or xA for short) is a measurement of the quality of the chance created by a player, based on the likelihood that a pass will become an assist.

This means that a pass that puts a forward through on goal will have a higher xA than a pass outside the 18-yard-box, even if both result in assists for the player.

It is important to note that the passing player still gets credit for the quality of their pass, whether or not the receiving player is able to convert it into a shot or goal. This means that it is not reliant on the finishing ability of the receiving player, but solely the quality of the pass provided.

Sources:
Opta Analyst - Great analysis of xG for and against. Tempo, pressing, pass sequences and also Zones of control. 
FotMob - xG and xGOT stats plus many more. Very detailed xG, literally showing each shot and the score associated to it!
Sofascore - Lot's of stats on the game. Attack momentum is a good indicator for performances. Also the 'Player average positions' is great.


Charlton Vs Exeter (3rd Oct 23)

As mentioned above, the detail on FotMob is awesome for xG breakdowns. They capture each shot and what score it is given for xG and xGOT. Link to this game here. You can use the Shot Map to scroll through each chance.


I'm probably going to add more to this later but just thought I'd start the ball rolling for now.

[Mundell]

Good that you’ve posted this aaronaldo And thanks for the FotMob link. I wasn’t aware of that site. 

One of the things I have consistently said on the “Lies, lies, damn lies and statistics” thread, and the reason I chose that name for it, is that it’s important to interpret this kind of data very carefully. To ensure we understand what a piece of data might be telling us and, equally importantly, what it’s not saying. It’s important not to over interpret too, drawing conclusions that might be true but don’t necessarily follow from the data. For all those reasons it’s good that you’ve started this thread with some definitions. And that we now have a thread to develop our understanding. 

In the spirit of ensuring we understand what the data is telling us, I have an observation about the XgF numbers on Opta Analyst and FotMob. In principle, we might expect these to be the same because FotMob references that it also uses Opta data, but on the surface at least they appear to be quite different. I think there’s a simple explanation and something a bit more nuanced and, potentially, quite debatable. 

As we know, Opta Analyst shows data for XgF from open play and XgF from set pieces separately. However, it excludes penalties. As a result, prior to last night, Opta showed us with eight actual goals from open play with none scored from set pieces. This was three goals short of the eleven we’d actually scored because they’d left out the three penalties we’d taken. The logic of excluding penalties from the Xg calculation (and hence from the goals actually scored comparator) is that they have such a high Xg that their inclusion can distort the comparison between teams, at least on a single match basis. A team that wins two penalties in a match hasn’t necessarily created the equivalent of two very good chances and their apparent ability to create chances and repeat that outcome might be overstated. 

By contrast, it’s clear that FotMob are including penalties in their XgF calculation and I think we can infer that the Xg per penalty is 0.765. In other words, they expect penalties to be converted about three times in four. En passant, I’d note that we may feel, intuitively, that this is a bit low, but even top players like Harry Kane struggle to get up to 0.8. Lots of penalties get missed. 

Anyway, this is all entirely reasonable provided we recognise that the Xg of teams that have won a lot of penalties is probably flattering them a bit, unless they can keep winning them at the same rate. 

However, I think there is something else going on which is a bit more contentious. Opta Analyst appear to have included last night’s penalty as a set piece goal and must, therefore, have included the Xg of the chance from the rebound in our Xg from set prices. I’m not sure I’d have done that. It’s a bit odd to include the chance from a penalty rebound when the penalty itself is not included. Fair enough though. 

What isn’t fair enough though is that FotMob appear to have double counted, i.e. they appear to have included both a penalty Xg of 0.765 and the Xg of May’s chance from the rebound. That’s simply wrong. 

In the grand scheme of things this is all relatively minor stuff obviously, but I do think the way analysts produce data tells us about their focus on doing ‘good work’ versus throwing out lots of information they hope to eventually commercialise. I’d be wary of reading anything into numbers we don’t understand. There’s a fair chance those producing them don’t really understand them either. 

BTW, I think post shot Xg is a really useful data point. Do you have any sources for that? 

I’ll look forward to your next post on this thread.

[aaronaldo]

Neglected this thread for a while, but check out this list of resources: www.ntt20.com/p/the-ntt20-research-bible?r=1l2n8x&utm_campaign=post&utm_medium=web

Really great list from the Not the Top 20 lot. Loads of great resources for anyone interested in different aspects and it explains why each are useful. I use Fotmob a lot to follow games. Great app for your mobile. The 'momentum' shows a little on how the game is going and the pressure being applied, which isn't always easy to tell from the standard live scores/text updates.