My issue with xG is that I don’t know the input for the metric so I can’t really say what it’s good for and not good for. Especially in small sample sizes.
But it’s a predictive metric so it will have been built/tested etc enough to be accepted by the entire football analytics community. The issue is how it’s used, not the metric itself.
If xG says that it gives 0.5 xG for a chance that’s because over a large enough sample size 50/100 times that chance results in a goal. That’s all it says and that’s all it can be used for. It gives you questions not answers.
I don’t know if it takes into account left or right footed chance for a right footed player, I don’t know if it takes into account form of the player shooting, I don’t know if it takes into account the quality of the GK…, there is countless other bits of missing context that people fail to consider when using it as a single game answer for why a game was won or lost.
And to be clear, it’s only ever used for single games in data editorial in the media. No serious analyst for a team should be quoting it as an answer for why a game was good/bad for a team.