Single Game xG is here to stay - is it useful?
/By Jared Young (@jaredeyoung)
It’s been documented for a while that Expected Goals (xG) is the best single metric for understanding performance of a soccer team and predicting it’s future. It outperforms possession, total shots ratio, goal difference, points scored, and other fun but inferior statistics. The word on Expected Goals has been slowly on the rise since ASA’s first model was released in 2013. For kicks, below is an indexed view of times the metric has been searched for in the United States since that grand moment in time. Searches reached an all-time high this past February.
source:trends.google.com
While Expected Goals is getting its due focus, it does take about seven games before it unveils its magical powers. A single game is predictive of absolutely nothing. Yet this very website as well as MLSSoccer.com show game level results with xG. And here’s ASA contributor Eliot McKinley pointing to one game’s xGD (Expected Goal Difference) as evidence that Nashville fans shouldn’t be discouraged by their first game loss:
In these cases Expected Goals is being used to describe something about a single game already played, not a game in the future. Not a lot of context has been provided about the usefulness of Expected Goals to analyze one game, so let’s add some to the discussion. I compiled the results of all MLS matches between 2011 and 2019 and examined the probability of win, draw or loss for various levels of Expected Goal differences. Turns out single game Expected Goals is reasonably well correlated with the outcome it suggests.
A look at average points across difference xGD outcomes shows that there is a generally linear correlation between xGD and actual points scored. There’s not much of a distinction if the home team has negative xGD between 0 and 1, but there’s a big drop off if they perform much worse. On the positive side, the better the home performance the more points they’re expected to earn. Here’s that same view looking at the distribution of outcomes.
The dark blue line is the distribution of the home team having that xGD in a single game. That a home team wins the xG battle by 2.0 or more happens 4% of the time, for example.
When Home xGD is less than -1.0 the home team wins just 21 percent of the time. When the xGD is greater than 2.0 the home team wins 78 percent of the time. Interestingly, the probability of drawing does not change much across the outcomes. Draws range between a 16 percent chance and 31 percent.
Let’s use the Nashville example to dig further. According to the ASA model, Nashville tallied 0.73 xG while Atlanta mustered just 0.27 xG, a difference of 0.46. That the home team “won” xGD by between 0.25 and 0.50 points happens 12% of the time. That’s roughly the median result for nine seasons in the analysis. Teams that pull that off win the game 51 percent of the time and draw 31 percent of the time, so the fact that they lost happens just one in five times. It is worth mulling that even a separation of half a goal in terms of xG only earns a team a coin flip when it comes to winning. Soccer just has a way of being unpredictable.
Mr. McKinley’s consolation to Nashville fans appears to be justified. Nashville performed roughly average at home against one of the East’s historically best sides, and while they lost, it’s a situation they will win half the time. Expected goals do tell an interesting story after a match. Now we just have to wait a half dozen more games to find out how good Nashville should expect to be.