xG in FM? What does this mean?

Hello reader, welcome to Get Goalside!. If you haven’t already seen it, here’s a video of an AI camera mistaking a bald linesman’s head for the ball.


*Electronic Arts voice*

Ex. pected. Goals. It’s in the game.

Ok, so this isn’t about EA Sports, but xG is in FM. A nice big announcement from Football Manager said they’d have the metric in the 2021 edition of the game. There’s a proper data company edge to it too, thanks to a partnership with Dutch company SciSports.

If you want to read a selection of ‘the stats community’’s reactions, you can check out the responses to my tweet when the news first broke:

In terms of gossip, there was also a reasonable amount of raised eyebrow when the very first screenshots came out (a few days before the announcement of the SciSports partnership) and the xG map colour scheme looked a lot like StatsBomb’s.

Caption

Ted Knutson explained away the similarity:

And the next video had a different colour scheme (as well as a nifty SciSports logo):

Caption

Share

It isn’t too surprising that the original mock-ups might have had a coincidental resemblance to StatsBomb’s work. Firstly, nerds love FM and nerds love StatsBomb. Secondly, named data companies have been in the game before, with STATS (now StatPerform after merging with Opta) being the ‘in-game data provider’ in previous iterations.

And now it’s SciSports’ turn (it seems that if you want the attention of Sports Interactive, who make FM, you need to be a company beginning with ‘S’. My employers Twenty3 must be kicking themselves at being alphabetically so close.)

It’s quite a sweet story. SciSports founder Giels Brouwer says in the press release,

I have always been a passionate Football Manager gamer and it is a great source of inspiration for our entire team. When we founded SciSports, we always said that our ultimate goal was to build Football Manager with real match data for pro football clubs.

But although the partnership has only just been announced, this hasn’t been a rushed relationship. I spoke to the company’s Chief Analytics Officer, Jan Van Haaren, about the partnership and how it had come about. “SciSports and Sports Interactive have a long-standing relationship that goes back many years,” he said, “Delegations of the two companies have visited each other's offices on multiple occasions throughout the past few years to explore potential collaborations.”

SciSports aren’t quite building the xG that will appear in-game though, the data company acting more in an advisory capacity to Sports Interactive. “Sports Interactive developed their expected goals model in-house using data that was produced by the Football Manager match engine,” Van Haaren explained.

“However, the SciSports and Sports Interactive data science and development teams met regularly to discuss the progress and challenges. SciSports' prior expertise with building advanced metrics such as expected goals helped the Sports Interactive team to identify information that is predictive of the outcome of a shot, and to improve the accuracy and robustness of the expected goals model.”

It sounds like an interesting challenge though. The apparent beauty of working on this kind of model in a game like Football Manager is how rich the data is, “such as access to the spatial locations and body orientations of all the players at all times,” Van Haaren said. However, “the computation of the expected goals values needs to be fast as many matches are simulated in the game while the computational resources might be restricted on some devices”.

The section of stats twitter that were wondering about Football Manager’s match engine in response to my tweet will also be interested in this next bit.

“Since the Football Manager match engine is not deterministic [it doesn’t just flip a coin and say ‘goal’ or ‘not goal’] and the outcomes of shots are influenced by many different factors,” Van Haaren explains, “the expected goals values do provide insights into the performances of teams like they do in real life.

“In addition, the expected goals values also help to further improve the realism of the Football Manager match engine. That is, over a large number of matches, the distribution of the expected goals values for real-world shots should be similar to the distribution of the expected goals values for the simulated shots.”

Neat.

If you’re an avid FM player you might have already seen the latest news on the match engine AI improvements that’ll come in for the 2021 edition of the game. I spoke to Van Haaren before that news dropped and didn’t know it was coming, and I’ll leave any speculation as to how linked the two are up to you, dear reader.

It all raises an interesting point about a new frontier of video game realism. I confess, one of the reasons why I spent more time on the AI improvements is that I don’t really play FM. A minor part of that, and something that might tempt me back if it was properly addressed, is the lack of women’s teams, leaving devoted fans to put databases together,

But part of it is just that I don’t enjoy it very much. Mostly because I’m not very good; a little bit because there’s a lot more to do than when I played it when I was younger (c. 2005-2012).

FIFA and Pro Evo have similar, although slightly different, problems. How much should they aim to reflect ‘real football’ and how much should they consciously diverge to make a fun game?

I’m not saying that xG will take the fun away at all, and actually think it would improve my enjoyment if I was able to work out whether my tactics were actually working or not. Let’s just hope that FM 2022 doesn’t involve the capability to do this:

Pass it on if you’ve enjoyed this newsletter, and subscribe if you think you’d like it again in future