A recent poll conducted on Twitter by a stats-based account asked users a simple question - do you know what Expected Goals means? 59% of those asked responded no.
It's a desponding response given that xG has been with us for quite some time now and puts any future developments into perspective - if people can't grasp something that is used reasonably regularly in analysis articles, or perhaps don't want to, then it doesn't bode well for anything more complicated.
But Expected Goals should be straightforward enough to understand, as should its limitations.
EXPECTED GOALS: DEFINITION?
Expected Goals, shortened to xG, measures the quality of a shot based on several variables, such as the angle of the shot and the distance from goal.
Some models are more sophisticated and include a broader range of inputs such as whether the effort was either a shot or a header (headers are, generally, lower quality opportunities), the phase of the attack, and whether the opportunity can be defined as a Big Chance.
The more shots that are provided to the model, the more accurate the model can be, but even very basic models will take up to 10,000 shots on goal and provide a percentage likelihood, based on the database of shots, of a chance from that position finding the net.
Of course, the more shots that are provided, the richer the database and, theoretically, the more accurate the % representation of a player scoring from that position on the field.
WHY IS EXPECTED GOALS USEFUL?
Goals and assists don't always work out. Some players can be unlucky and miss a number of chances they would normally be expected to score, while others can overperform their Expected Goals total by going on a run of scoring with very few shots on goal.
A good expected goals model shouldn't be outperformed on a consistent basis by a player. Therefore if a player is underperforming from a goals scored perspective but has a high xG total, we shouldn't worry too much, because he is getting into the positions where he can score goals.
If a player is overperforming, we should note that the player has been particularly clinical in the short term, but be aware that over time his goals scored figure will likely realign with his expected goals total.
OVER WHAT TIME PERIOD IS XG USEFUL?
Expected Goals is most useful over an extended period of time - a season, or in most cases, longer. We can get a clearer indication of a player's propensity to get into good scoring positions - ie positions where his xG number is reasonably high - over time, and use this as a clearer benchmark than simply their goals totals.
Generally the situation that you don't want your forward to be in is not scoring goals, and also having a low xG. This means not only are they not scoring, they aren't getting into positions where they are likely to score, either.
DOES LIONEL MESSI LIKE XG?
Lionel Messi should be excluded from most statistical conversations simply because he is an extreme outlier in so many, so his is not a performance to be benchmarked against.
But there have been reports in recent seasons that Messi is taking a more keen interest in xG from the respect of shooting less, but shooting from more optimum positions. And across the last three seasons his numbers do reflect this shift. It may become a more common thing as players begin to realise that shots from outside the area are often a one in 50 (0.02 xG) or 1 in 100 chance (0.01 xG).
EXPECTED GOALS LIMITATIONS
Expected Goals isn't perfect, for a number of reasons. First of all there are many different models available, all providing different numbers and approaches to their calculations, which can be confusing. And some of the inferior ones don't have enough data behind them in order to be as accurate as others.
The best xG models consider many elements of the play including shot position, angle to goal, the type of assist, and the body part used to connect with the chance. They also consider the phase of play - if a player is counterattacking it's likely he has less defenders between him and the goal when he chooses to shoot.
But the lack of comprehensive player tracking data across all leagues makes it difficult for models to assess how many defenders are between the ball and the goal when a player looks to shoot. From this perspective, xG can sometimes overrate a more difficult opportunity, particularly when close to the goal. As with every statistic it's worth analysing the video of the chances themselves to give a clearer indication of what the player was facing.