Do xGoals win games?

As a football fan and also a keen poker player I’m often frustrated by the high level of results orientated thinking that seems to dominate football punditry.

In poker if you hold two Aces and you go all in against an opponent holding two Kings you will lose the hand 20% of the time. Losing that specific hand does not mean that next time you're dealt Aces you should fold them.

In football however, if Man City were to dominate Huddersfield Town for 89 minutes without scoring and then lost to a last minute counter attack, the typical response from pundits and fans would be to criticise City’s inept strikers and heap praise on the brave Huddersfield defence.

The reality is that there is a level of variance (luck) present in any single football match just as there is in any single poker hand. In the long run the best team and hand will win and as such pundits and fans shouldn’t use the result of a single game as evidence of a team’s long term prospects.

Football in general is beginning to wake up to the power of statistics and in 2017 the ‘Expected Goals’ (xG) Opta stat was added to Match of the Day’s weekly analysis of the Premier League. xG attempts to measure what ‘should’ have happened during a match based on the number and quality of the chances created by either team. At the end of each match each team receives an xG rating and the differential between these figures is used to generate a new stat called ‘Expected Points’ (xPoints).

The introduction of xG had the potential to reduce results orientated thinking however, so far it has resulted in a somewhat underwhelming reaction from pundits…

“He’s the first person I’ve ever heard take any notice of expected goals, which must be the most useless stat in the history of football! What does it tell you? The game’s finished 3-1, why do you show expected goals afterwards?” Jeff Stelling on Arsene Wenger, 2017

In addition to this TalkSport produce a weekly online article to highlight “how stats experts think the Premier League table SHOULD look” based on xPoints and presumably also to revel in the inevitable differences caused by short term variance in results.

On the surface this scepticism seems reasonable as a quick glance at a table comparing xPoints from the 2018/19 season so far with the actual points accumulated shows a huge delta for many EPL sides*…

*13 sides selected on basis of being consistently present in the EPL since at least 2015/16

On average xPoints reflects a team’s actual points to a 14% margin of error which is significant enough to wildly skew the positions in the table.

But as with all statistics subject to variance, in order to measure whether xPoints is actually useless, we have to try to look at a large enough sample size to counteract the variance inherent in the game. I’ve therefore, collated all of the xPoints from the Premier League since the 2015/16 season (as far back as I could find) up to match 21 in 2018/19 and compared them to the actual points totals achieved. I’ve done this for all 13 teams who have been consistently competing in the Premier League during this period and the results are below…

What this data shows is that on average over the course of three and half seasons xPoints reflects a team’s actual points total to a 7% margin of error. xPoints also reflects the exact actual points total for both Man City and Leicester City during this 135 match period.

There are however, some significant outliers that require further analysis as the gap between actual points and xPoints is 25 or higher and is therefore, unlikely to ‘correct’ within a single season…

Over Achievers

• Man Utd have significantly over-achieved in terms of their actual points accumulation v xPoints. This is perhaps due to the impact of World class Goalkeeper David de Gea and/or Mourinho’s traditionally defensive but efficient brand of football.
• Tottenham have also significantly over-achieved. This is possibly due to the combination of having one of the strongest defences during this period and a highly efficient striker in Harry Kane, who scored 98 actual goals compared to 82 xG.

 Under Achievers

• Southampton have significantly under-achieved during this period. This could be due to their struggles in finding a prolific goal-scorer, the top scorer at the club during these season’s is Sadio Mane with 15 goals and the club only averaged 1.17 goals per game.
• Crystal Palace have also significantly under-achieved. This seems like a similar situation to Southampton’s with Christian Benteke’s 15 goals in 2016/17 the highest individual total and a club average of just 1.13 goals per game.

Overall this data suggests that xPoints is a useful statistic over a significant sample size. It would therefore, be wise for football clubs, pundits and fans to not read too much into games where the actual result significantly deviates from xPoints. Whilst unfortunately it seems that a single season is not a large enough sample size to ensure that each club finishes where they ‘deserve’ to, the best long term strategy is likely to be to play in a manner that maximises xPoints.

It will be interesting to repeat this analysis in a few years time to see whether xPoints accuracy will continue to increase until the margin of error is close to zero. Alternatively it may be that other factors such as owning an especially efficient striker or strong goalkeeper e.g. Kane or De Gea are significant enough to enable a club to consistently outperform their xPoints total.