If stats are to be included in the game, respect it in your engine.

[Scamander]

My team is underperforming xGD by 0.5 per game, which is literally even worse than Brighton last season. Whilst the performance of Brighton in real life can be attributed to the lack of quality in their squad, my team should have easily above average finishing skills in the league. And no, I didn't put an outfield player in goal.

Nobody enjoys playing saves like this. If you want to increase the difficulty of the game, do it by having the AI employ better tactics. Not by having your engine **** over the player for reasons completely outside of the player's control.

[HUNT3R]

Firstly, there's nothing artificial in the game to 'increase the difficulty' as you claim. The ME also cannot distinguish between human and AI, so there's no bias there either.

It's quite possible to over-perform vs your xG. You're going to have to look at how you set up, the type of chances you're creating (both in terms of where, how much space and whether it's a shot or header etc) and also the player(s) the chances fall to. 

In FM20 and FM21, I massively over-performed against my xG. I'm even doing that right now, even though my team is one of the worst in the league and sitting in 13th out of 15. I'm sitting at 1.21 goals/match vs 1.13 xG/match.

[Scamander]

2 hours ago, HUNT3R said:

Firstly, there's nothing artificial in the game to 'increase the difficulty' as you claim. The ME also cannot distinguish between human and AI, so there's no bias there either.

It's quite possible to over-perform vs your xG. You're going to have to look at how you set up, the type of chances you're creating (both in terms of where, how much space and whether it's a shot or header etc) and also the player(s) the chances fall to. 

In FM20 and FM21, I massively over-performed against my xG. I'm even doing that right now, even though my team is one of the worst in the league and sitting in 13th out of 15. I'm sitting at 1.21 goals/match vs 1.13 xG/match.

If a team can consistently overperform xG, then there's either something wrong with the engine, the xG model in the game, or the team consists of multiple Messis. The types of chances really should have very little to do with xG conversion rate, as that is part of the consideration within all mature xG models; if the game is letting in a 0.05xG header 10% of the time on average and a 0.05xG volley 2% of the time, then again either the engine or the xG model is faulty.

[HUNT3R]

6 minutes ago, Scamander said:

If a team can consistently overperform xG, then there's either something wrong with the engine, the xG model in the game, or the team consists of multiple Messis. The types of chances really should have very little to do with xG conversion rate, as that is part of the consideration within all mature xG models; if the game is letting in a 0.05xG header 10% of the time on average and a 0.05xG volley 2% of the time, then again either the engine or the xG model is faulty.

https://understat.com/league/EPL

Just as an example here, as you can see, teams can be better or worse than their xG. Sometimes by quite a lot. Look at Chelsea, both in terms of goals scored and goals conceded, for instance.

 

In addition, the types of chances do matter. Headers are low % compared to shots from the same area. That's not the only thing though as your players (who the chances fall to) matter here as well. For example, if you're creating a type of chance that often means a player needs to shoot with his left foot, but he's right footed, he's going to struggle converting it (correct me if I'm wrong, but footedness isn't a factor in xG?) so it's important to analyse chances.

[Rashidi]

 

There is a variance between actual goals and xG all the the time, this variance is used to analyse whether a team is under performing/out performing or meeting expectations. 

While we should not expect every team to outperform xG, a typical variance of around 30% is reasonable to expect between actual goals scored and xG. However if you are not seeing any team underperform, then perhaps you should raise it as a bug