Stats, xG, etc.

[paul_e]

It's very easy to ignore if you don't like it. I'd imagine scouts and managers use it to some degree, along with numerous other stats and then the more traditional method of actually watching players once they have identified targets more specifically.

This is it. It's clearly a stat which has some uses, but almost never tells the whole story and certainly never in isolation. Like any metric, in fact.

... and the people who made it, along with the people thy made it for, are all aware of all of those limitations.

[German James]

It's very easy to ignore if you don't like it. I'd imagine scouts and managers use it to some degree, along with numerous other stats and then the more traditional method of actually watching players once they have identified targets more specifically.

This is it. It's clearly a stat which has some uses, but almost never tells the whole story and certainly never in isolation. Like any metric, in fact.

... and the people who made it, along with the people thy made it for, are all aware of all of those limitations.

Fair enough. Why is it suddenly "a thing"?

[Monty]

It's very easy to ignore if you don't like it. I'd imagine scouts and managers use it to some degree, along with numerous other stats and then the more traditional method of actually watching players once they have identified targets more specifically.

This is it. It's clearly a stat which has some uses, but almost never tells the whole story and certainly never in isolation. Like any metric, in fact.

... and the people who made it, along with the people thy made it for, are all aware of all of those limitations.

Fair enough. Why is it suddenly "a thing"?

Podcast twats, mostly.

[paul_e]

Yep, it became part of the opta stat pack and people who didn't really understand it started acting like it was the greatest stat ever and basing their entire output on it.

[German James]

Got it! Thanks. I expect it will blow over.

[lovejoy]

I also think some xG measures accumulate shorting the same phase of attack thereby inflating the xG regardless of the fact that you couldn’t score twice in the same move.

[Chris Harte]

I also think some xG measures accumulate shorting the same phase of attack thereby inflating the xG regardless of the fact that you couldn’t score twice in the same move.

I think there was an example of that recently involving Newcastle (possibly against us) where two players tried and failed separately to get on the end of a cross. They each scored something like 0.4 xG. Like you say, it's daft.

[paul_e]

That's only wrong if you then make the incorrect assertion that xG can all be added up and give you a 'score' of how many goals you actually deserved in the game, which is a perfect example of how it's being used wrong. If you say in that move they had 2 good chances to score but missed both of them that would reflect reality and that's what xG shows as well, it's just that it becomes nonsense when you then extend that to "they should have 1.2 goals for those chances".

In addition, even if adding chances up was somehow a reasonable use of the data using it to say a game should've been x-y instead of a-b is a complete misunderstanding of how probability works. 10 chances of 10% is not the same as 1 chance of 100%.

Moving away from sport for a minute, statistically 10 events with a 10% chance means, overall, you'd have about a 65% chance of a positive and you'd need about 41-42 chances of that to get to 99% certainty. However 10 20% chances jumps up to 89% and a 25% chance gets you to a 95% chance from those 10. This means that, from a purely statistical perspective, the stat is being completely misused even by the act of adding them up rather than just rating each individual shot.

However that's only really relevant if your underlying percentages are accurate. The problem xG has here is that, even with all the data they've given it, the sheer amount of variables used to create the value means that that accuracy is almost impossible to guarantee with any certainty. There are situation where the certainty jumps significantly though, for example penalties get an xG of 0.79 and that's because 79% of them get scored and they're such a uniform event that you can have confidence in it.

To use an extreme example the Beckham lob from halfway would clearly have a 0.01 xG (which is the lowest you can have) but that's not really based on anything because in the history of the sport there won't be 100 shots from that spot so even removing all other variables you're just guessing based on the distance from the goal and some arbitrary probability bandings.

Sorry, I'll shut up before I get really stat nerdy.

[Beard82]

Most of my career has been based around data. I run a small agency, and the majority of the digital work we do is related to data in some way.

Data is amazing - but it needs to be used correctly.  In business as in football "non-key" data (key = Goals / Points, Sales / Profit) is best used to inform better decisions / strategy / training etc.

You give yourself a better chance of success by understanding what the data is telling you but often its about understand why stats and results dont match expectations.  I.e. What is your best sales man doing differentk or  if a player was expected to score 2 goals but didnt - why and what can be learnt from it.  Similarly, if you keep outperforming a metric - why and what can be learnt from it.

That said - typically data is misued into validating a view - rather than informing it