My conclusion on xG is that it is actually really good for assessing team defensive and offensive overall performance (ie., xG for and xG against)… to me, it isn’t nearly that good at analyzing individuals because so many of the variables that go into it are related to other people, unless we are talking about mazy-run dribbling exploits that lead to a shot after beating 5 people.
For example, if a player happens to get into the box a lot and his teammates find him, he will have xG through the roof. If he then underperforms his xG, or over performs it, how should that be interpreted? On one hand, you might argue that the player is tricky and finding spaces in crowded zones, doing hard work to find himself open enough to receive and shoot, so a lower conversion rate is not punished as much. On the other hand, the player might be considered wasteful if he underperforms the xG.
Probably more importantly, one shot from inside the box close in can contribute a LOT to xG, so those little single events (again, rare event stats) can have massive impact on a player’s single metric for a game and a little less so for a team’s full xG (make of that what you will). Also, it exclusively measures attempts at goal - nothing else… any offensive or defensive contributions that didn’t lead directly to an attempt at goal are not considered.
In general, I think the model is pretty good predictor over time of a team’s performance though - Nice being the obvious anomaly.
One other thing - xG is not a single algorithm or measure - if you look, there are numerous models built around the concept. Some of them correlate well with team goals, others less so.
What I would find very interesting is if they can expand on it to look more at an xG or xD (delta in goals) as models of showing a players’ total contribution. I suppose one could do a crude measure of team effectiveness to when a player is on the pitch and not, but that is harder given the constantly changing conditions around those variables.