The Growth of Expected Goals
Expected Goals has found its way from blogs and articles to mainstream media coverage over the past eighteen months. Yet like all new metrics one can get carried away with the advantages to the user and forget about the negative elements.
What is ‘Expected Goals’
Expected Goals (xG) is all about the ‘quality of the chances’. How likely or ‘expected’ is a shot to result in a goal when compared to a big sample of shots taken from the same position on the pitch.
Expected Goals is usually shown as a number between 0-1; ‘0’ being a definite miss, and ‘1’ being a definite goal. You can also present the figure in percentage terms, so shot on goal with 0.3 xG for example, would mean a shot with a 30% scoring chance
One of the big reasons why Expected Goals is so popular is that it can be used by football teams, pundits, and the general public. Teams and managers can analyse players and performance. Pundits can use it as reference point for matches, and the general pubic can use it as a betting tool because it has better predictive qualities than total shots and so on.
Interpreting the Data
Many people don’t like statistics ‘intruding’ into their favourite sports and there is a lot to be said for simply ‘watching the game’. But if you are a ‘professional’ or even a casual bettor, then it would be irresponsible not to take advantage of information that is quite freely available.
There are, however, some limitations when using xG. You need to be aware of these, especially if you are planning to use it to bet on football.
Choose your Model Wisely
Which type of Expected Goal model is best? Simple? Complex? I have mostly found that simplicity is often best with stats and systems. You can of course go beyond the basic analysis of ‘shot location’ and ‘goal chance percentage’ – adding in factors such as ‘defensive pressures’, ‘anatomy’ used (headed goals) ‘assist type’, ‘goalkeeper positioning’ and so on. Yet often these do not particularly add to the predictive qualities of the model.
What xG Doesn’t Tell Us
Most xG models don’t allow for the quality of individual players. Most models are built around a large sample of shots, with the final output presented as an average.
This does not take into account individual player information, means that if a player like Sergio Aguero (striker) and Nacho Monreal (defender) were both in the same position on the pitch and taking on the same shot – the same xG figure would be applied to both chances, even though it’s clear that Aguero would be the more likely to score.
Similarly, the goalkeeper is also important, and it’s another parameter that xG doesn’t allow for. Shooting against David De Gea or Alisson Becker is a different proposition to shooting against a more average keeper. Yet xG won’t generally differentiate.
How to use expected goals
One of the big issues is that xG is not effective for a single game or a just a ‘couple of games’. Football is always ‘changing’ in terms of the teams and the action, who is on the field? Has there been a sending off? How much time is left in the game? What type of game is it (league, cup) ?
Xg is much more effective over longer periods of time. But timing is important too. Often it is better to wait out the early parts of a season, at which point the form and data will be more comprehensive – around week 6 or 7.
There are also times when you shouldn’t just ‘run with xG’ even if the timing seems right. Football teams can do well or decline for a whole host of reasons at points during the season, managerial changes, formation changes, key injuries, unrest in the camp. Looking at a side’s under or over performance becomes more important.
So, like any statistic, xG can be useful but can also quickly become redundant.
xG is certainly a beneficial metric for many people, as long as you acknowledge the limitations as well as benefits. It’s not just about the model or data that you analyse, it is also about interpreting it correctly.Acknowledging the points in this article should help you to use Expected Goals more effectively.